A laboratory report is a document that summarizes and analyzes a scientific experiment carried out in the laboratory. Its purpose is twofold:
To provide an accurate record of the activities carried out.
To offer a critical analysis of the results obtained, relating them to the relevant theoretical principles.
A well-written report should be clear, complete, and well organized, and should include data, graphs, and explanations that make it possible to reproduce the experiment reliably. This not only helps verify the results but also contributes to a better understanding of the scientific method.
The scientific method, introduced by Galileo Galilei, is at the foundation of laboratory reports and modern science. Galileo, considered the father of the experimental method, revolutionized scientific knowledge through a systematic approach that combined observation, experimentation, and mathematical analysis.
In a lab report, this method is reflected in the description of every stage of the experiment: from the initial observation to the formulation of hypotheses, through experimental testing and data interpretation. By following the Galilean approach, a rigorous and reproducible process can be ensured.
Physics experiments focus on the study of natural phenomena, such as motion, light, sound, and electricity, in order to understand the fundamental laws of nature. Through the scientific method, we begin with observations of the real world, formulate hypotheses, and test them through experiments.
For example, we might study how the velocity of a falling object changes or analyze the behavior of an electric current in a circuit.
These experiments, often carried out using instruments such as stopwatches, balances, or multimeters, allow us to connect theoretical concepts with experimental data, developing a deep and rigorous understanding of the physical world.
The scientific method, therefore, developed by Galileo Galilei, is at the heart of every laboratory report because it provides a rigorous and systematic approach to collecting and analyzing data. It can be divided into several fundamental stages:
Observation and formulation of the problem: Galileo asked precise questions based on observable phenomena.
Hypothesis: Following his example, rational explanations are formulated based on existing knowledge.
Experimentation: As Galileo demonstrated through his experiments, it is essential to design reproducible tests.
Data analysis: Galileo analyzed results mathematically, comparing them with theoretical models.
Conclusion: The results support or refute the hypothesis and lead to new questions.
A laboratory report documents each stage with clarity and precision, making the experiment reproducible, as Galileo advocated for the advancement of science.
When writing laboratory reports, it is a good idea to start using tools that may be useful in our future careers, especially Excel, as the meme on the right aptly illustrates.
Excel is the tool of choice in many professional and business environments. It is ideal for creating tables and graphs and for beginning to work with data in a systematic way. It is an extremely versatile tool, used in fields such as mechanical engineering, logistics, computer science, finance, and many others.
A key feature in lab reports is curve fitting, which allows us to find the mathematical function that best describes a set of experimental data.
While working with these tools, you will often encounter three terms that are important not to confuse:
Curve Fitting: this is the overall goal, which means adapting a mathematical function to fit our data points.
Linearization: this is the algebraic strategy we apply (e.g., squaring a variable) to transform a curved relationship into a straight line.
Linear Regression: this is the statistical calculation that Excel or Python performs on our data to find the optimal "line of best fit" and give us the exact slope.
For further information, I recommend the following links:
An example of how to set up a best-fit curve in Excel using experimental data is available at this link (error bars work in Microsoft Excel but not in Google Sheets).
To create lab reports, we can use digital tools that allow us not only to organize experimental data, but also to process, graph, and interpret it using scientific analysis tools.
For this purpose, I am providing a guided template for creating physics lab reports with Python (a programming language) and Google Colab (cloud-based - introductory video tutorial here) or Jupyter Notebook (local app - introductory video tutorial here), designed for high school students (especially useful for technical institutes or scientific high schools).
This is a general and customizable guide, to be used as a starting point for all lab reports.
Introductory guide to Jupyter Notebook and Google Colab: what they are, how they work, how to use text and code cells, and how to save your work.
Introduction to Python and scientific libraries: NumPy for numerical calculations, Pandas for organizing data into tables, Matplotlib for creating graphs, and SciPy for fitting procedures.
Organization of experimental data: how to enter measurements, manage uncertainties, and build organized tables.
Graphs and curve fitting: examples of experimental data representation, linear regression, and mathematical model fitting.
Report structure: a general outline with the fundamental sections of a scientific report, accompanied by instructions on what to include in each part.
Download the .ipynb file from the button below.
Open Google Colab and load the notebook via File → Upload notebook.
Save a copy to your Google Drive via File → Save a copy in Drive.
Follow the instructions contained in the text cells, replace the demo data with your measurements, and adapt the structure to the specific lab experiment.
Delete the explanations and sections you do not need in the final report, always keeping the necessary scientific content.
The notebook uses Python as an analysis tool. It can be reused for various lab experiments, adapting it each time to the characteristics of the experiment.
The sections with a gray background contain advanced concepts, so if you are looking for a simpler treatment, focus only on the sections with a light background!
Every measuring instrument is an object with specific characteristics and is therefore subject to measurement limitations, which can be divided into three main categories:
Instrumental Limitations: These depend on the specific instrument used. Their main characteristics are:
Range: The measurement interval allowed by the instrument, from the minimum to the maximum measurable value;
Resolution: The smallest discrete increment (step) that the instrument can display or indicate (e.g., 1 mm on a ruler);
Sensitivity: How "strong" the instrument's response is to a very small (infinitesimal) change in the measured quantity (it can be viewed as the slope of the instrument's calibration curve).
Note: High school textbooks often refer to sensitivity as "the smallest variation the instrument can measure," although the technically correct term in that context would be resolution. This is done for simplicity: it is more direct to say "how small I can measure" without getting into technical distinctions. Furthermore, for the simple instruments we use in school—like a ruler or a digital scale—sensitivity and resolution often coincide numerically, so this simplification works well in introductory approaches.
Random Errors (Type A Uncertainty): These errors arise from unpredictable variations in the measurement technique or the experimental environment (e.g., a trembling hand, the viewing angle when reading a ruler, or temperature fluctuations). Since random effects tend to cancel out when averaging numerous measurements, they can be characterized statistically to obtain an estimate of the dispersion of the measured values.
Systematic Errors (Type B Uncertainty): These errors stem from constant factors during the measurement process, such as incorrect instrument calibration. If, for example, a scale is improperly zeroed, all mass measurements will be systematically skewed. Systematic errors are ideally identified and corrected rather than included in the statistical uncertainty, as they cause a constant, predictable shift from the true value.
Therefore, for every measurement taken, we must report the corresponding uncertainties in our lab report. Below are the possible cases that may occur.
For an in-depth look at error theory, please refer to this link from the Penn University Physics Department.
When we take a single measurement of a physical quantity x, the resolution (or least count) of the instrument defines the smallest readable discrete increment and determines the absolute uncertainty of the measurement (Type B component).
Uncertainty is an estimate of the dispersion of the measured values and is distinct from error, which represents the difference between the measured value and the true (often unknown) value. For example, if a ruler has a sensitivity (least count) of 1 mm, the absolute uncertainty of the measurement will be ±1 mm, and we indicate it with the Greek symbol Δ before the measurement x:
Now suppose we measure the length of a smartphone (120 mm) and, with the same instrument, the side of a room (5000 mm), obtaining an absolute uncertainty of 1 mm in both cases. It is evident that the impact of this uncertainty is very different in the two cases: 1 mm out of 120 mm represents a significant fraction of the measurement, whereas out of 5000 mm it is almost negligible. To correctly evaluate the precision of the measurement, we introduce the relative uncertainty, defined as the ratio between the absolute uncertainty and the measured value itself. In this way, despite having the same absolute uncertainty value, the smartphone has a higher relative uncertainty than the room, highlighting how relative uncertainty provides a more meaningful indication of precision relative to the size of the measured quantity.
In non-specialized contexts, the term "relative error" is used interchangeably with "relative uncertainty," partly because knowing the true relative error would require knowing the true value of the measured quantity, which is often unavailable in practice.
This index provides a dimensionless measure of precision: a low value indicates that repeated measurements deviate little from one another, and are therefore very precise. However, it is important to distinguish between precision and accuracy.
Precision refers to the consistency of the results: repeated measurements that cluster tightly around a mean value indicate high precision, regardless of whether that mean value is close to the true value or not. Accuracy, on the other hand, expresses how close the mean value of a series of measurements is to the accepted or true value of the quantity. For example, if we measure the length of a sheet of paper and get 11.1 in, 11.2 in, and 10.9 in, the measurements are precise (the values are close to each other) and, since they average close to 11.0 in, they are also accurate. In other cases, highly precise measurements (very similar values) can still be inaccurate if the mean value is far from the true value, indicating a systematic error.
In summary, while relative uncertainty is a good indicator of measurement precision, accuracy is evaluated by comparing the measured mean value to a known reference value. If the known value is identified by the center point of the targets below, we have the following cases:
Low precision and accuracy
Low precision, high accuracy
Alta High precision, low accuracy
High precision, high accuracy
If we measure a quantity x multiple times to reduce the random errors made during the measurement process, its final value will be given by:
where the expected value x̄ of the measurement is the arithmetic mean of the N measurements:
while for the absolute uncertainty Δx of the measurement, we consider the larger value between the sensitivity (or resolution) and the maximum half-range (semi-dispersion) of the values, which is calculated as half the difference between the maximum and minimum measured values:
When we repeat a measurement N times, each measurement fluctuates around the true value due to random errors. The arithmetic mean x̄ of the measurements provides a more reliable estimate of the measured quantity, but we must also quantify the data dispersion (how much individual values deviate from the mean). Statistical analysis serves to quantify how much we can trust this mean.
The maximum error (maximum half-range) provides a rough approximation because it is highly sensitive to any outliers.
The standard deviation (or root-mean-square deviation) is instead a more reliable measure because it considers all measurements, weighing the squared deviations from the mean.
Bessel's Correction: We use (N-1), called degrees of freedom, instead of N. This corrects the bias created when using a limited data sample to estimate the true standard deviation of the entire population. Dividing by (N-1) provides the best unbiased estimate of the population variance, which is especially critical when N is small.
This formula derives directly from the definition of variance in statistics.
The standard deviation, denoted by σ, is the main index of how individual measurements disperse around their mean value. In simple terms, it tells us how "spread out" our data is.
A small σ indicates that the measurements are very close and clustered around the mean. This is a sign of high precision in the measurements.
A large σ indicates that the measurements are very spread out and far from the mean, revealing lower precision.
Therefore, σ describes the "width" of the distribution of the individual measurements taken.
The error associated with the mean of the measurements is not σ, rather the standard error of the mean σm:
This formula shows us that as the number of measurements N increases, the error on the mean decreases: the more data we collect, the more reliable our estimate of the mean becomes.
The distribution of the means is Gaussian. We can say with confidence that the interval x̄ ± σm has a 68% probability of containing the true value.
The Central Limit Theorem is the reason why this statistics works - it basically states that:
"No matter what the probability distribution of the individual measurements is (it might not even be Gaussian), the distribution of the means calculated over sufficiently large samples will always tend to follow a Normal (Gaussian) distribution."
This theorem allows us to use the standard error of the mean σm to build confidence intervals (CI), which tell us how confident we are in a given measurement. We have that:
The interval x̄ ± 1σm has a 68.3% probability of containing the true value.
The interval x̄ ± 2σm has a 95.4% probability of containing the true value.
The interval x̄ ± 3σm has a 99.7% probability of containing the true value.
However, this theorem is strictly valid only for an "adequately large" N (typically for N > 30).
❗ Limited data - pay attention to the number of measurements N!
The formulas for the mean (x̄) and the standard error of the mean (σm) are the most correct estimates for the true value and its uncertainty, for any number of measurements N. However...
With a limited number of measurements (typically N < 30), the interpretation of this uncertainty changes. We can no longer assume Gaussian statistics (the standard "bell curve"). To build a rigorous confidence interval, one should use the Student's t-distribution, which provides wider intervals to account for the scarcity of data.
In other words, with only a few data points available, it is like trying to guess the shape of a drawing by seeing only a few fragments of it. Our small collection of measurements might not have captured the true variability of the phenomenon. As a result, we cannot safely apply the powerful rules of statistics based on the "bell curve", known as the Gaussian distribution, but we must use a distribution called Student's t where the confidence interval becomes x̄ +- t * σm. Here, t is a factor (taken from specific tables) that depends on N (specifically, on the degrees of freedom N-1) and the chosen confidence level. For small N, t is greater than 1 for standard confidence levels, making the confidence interval wider to reflect our greater uncertainty.
When a quantity is determined from direct measurements (for example, calculating average speed using distance traveled and time taken), there are different strategies to estimate the uncertainty of the final result:
Single Measurement and Error Propagation (basic approach): we perform a single set of measurements of the quantities that make up our derived quantity (for example, one value for Δs and one for Δt), and to obtain its uncertainty we use the rules of error propagation (next section).
Simple Mean of Repeated Measurements (simplified statistical approach): we repeat the experiment multiple times for different sets of quantities that make up the derived quantity (e.g., different Δs and Δt to calculate speed) and we calculate the value of the derived quantity for each pair (v1, v2, v3, …). We then calculate the mean value as the arithmetic mean of the results and estimate its uncertainty directly from the statistical dispersion of the derived quantity values (e.g., maximum semi-dispersion or standard error of the mean), ignoring the uncertainties propagated on each individual value.
Combined Propagation and Statistical Analysis with Weighted Mean (rigorous approach): we repeat the experiment multiple times and for each trial we calculate both the value of the derived quantity and its uncertainty through error propagation (v1±δv1, v2±δv2, …).
The results are then combined using a weighted mean, where the most precise measurements (with the smallest δvi) have a greater influence.
The final uncertainty will be smaller than the smallest of the individual uncertainties, because combining more data increases our confidence in the result.
The weight of each measurement is calculated as the inverse square of its uncertainty: This represents its reliability or the amount of "information" it carries.
The fundamental quantity that measures statistical dispersion is variance, which is the square of the standard deviation (in our case, the square of δvi). The reliability of a measurement is therefore inversely proportional to its variance.
Using the inverse square ensures that clearly better measurements dominate the final result much more decisively, as is appropriate.
The weighted mean value is calculated as:
The uncertainty of the weighted mean:
For further details, we suggest consulting John R. Taylor, "An Introduction to Error Analysis: The Study of Uncertainties in Physical Measurements", Chapter 7 "The Weighted Average".
Another reference text is P. R. Bevington, "Data Reduction and Error Analysis".
When calculating a derived quantity (like density from mass and volume), you go through a multi-step process. One of the fundamental rules to avoid compromising the quality of the final result is: never round the values of intermediate calculations.
Every time you round a number, you "chop off" digits, losing information. This introduces a small inaccuracy, called rounding error. If this is done multiple times during the same calculation procedure, these small errors can add up and accumulate, leading to a final result that is noticeably different from the correct one and less precise than it should be.
Rule: Round Only at the End
When doing calculations by hand, it is admittedly inconvenient to keep all the decimal places: for this reason, we tend to truncate or round immediately, but we must be aware that by doing so we accumulate a "premature rounding error" that can become significant in the final result. If we use Excel (or any spreadsheet), we should keep all the digits the program gives us and not round until the very end.
Apply the significant figures rule only to the final uncertainty
Uncertainty is rounded to 1 significant figure, but if that first digit is 1 or 2, you can keep two figures (e.g., from 0.018 → 0.018; from 0.058 → 0.06).
The idea of rounding the uncertainty to a single significant figure (or at most two, if the first is 1 or 2) stems from the fact that the uncertainty itself is an estimate of the possible error, and not a fixed value "measured with precision". Adding extra significant figures would give the illusion of knowing the error itself better, while in reality:
The uncertainty has its own "uncertainty": when you estimate the standard deviation from a finite sample, the result can vary significantly (even up to 10–20%) if you slightly change the data.
Giving too many digits has no practical meaning: saying ±0.003724 g suggests we know the error down to the millionth of a gram, whereas the statistical estimate of σ with a few dozen measurements is not that reliable.
International convention: laboratory guidelines dictate limiting uncertainty to 1 significant figure, reserving 2 only if the first digit is "1" (to avoid too large of a relative jump).
In practice: if you calculate an uncertainty of 0.003724, you round it to ±0.004; if you get 0.058, it becomes ±0.06; but if it is 0.015 (starts with 1), you can keep ±0.015 or round to ±0.02, depending on the specific rule adopted. This ensures that the reported uncertainty itself remains "uncertain" in the right way, without making your estimate seem more precise than it actually is.
Once the uncertainty is rounded (for example to ±0.02), you know that your measurement is only reliable up to the hundredths place. Consequently, you must also express the measured value with the same precision:
If the "raw" result of the calculation is 12.3451237, it is rounded to the hundredths, i.e., to 12.35, because the found uncertainty is ±0.02.
This way, value and uncertainty are consistent: x = (12.35 ± 0.02). Here 0.02 indicates the precision of the last decimal (hundredth), and it would make no sense to report the value with more digits than those with which the error is actually known.
If a quantity Z is obtained by adding or subtracting two measurements A and B:
The absolute error is given by the sum of the absolute errors of the individual measurements:
In reality, the linear sum approximation of errors is only valid if the errors are systematic or strongly correlated.
Sometimes, however, errors are independent and random: while one is positive, the other is negative, so they do not always add up in the same direction. In statistics, when two independent variables are added together, their variances sum:
If the errors are independent, their fluctuations can partially cancel each other out, so the total error is less than the direct sum.
Since the absolute error is the standard deviation, we obtain the sum in quadrature of the errors:
If a quantity Z is obtained by multiplying or dividing two measurements:
The relative error is given by the sum of the relative errors of the individual measurements:
The formula is obtained from the partial derivatives of the function Z, specifically by applying logarithmic differentiation which allows simplifying the calculation of the derivative:
A special case is the multiplication of a quantity by a fixed number k:
The use of partial derivatives in error propagation has a profound meaning: it serves to estimate how a small variation in an input variable affects the final result.
Imagine that a quantity 𝑍 depends on several measured variables 𝐴, 𝐵, 𝐶, …, each with its own error 𝜎𝐴, 𝜎𝐵, 𝜎𝐶 ,…
If 𝑍=𝑓(𝐴,𝐵,𝐶), its value will change if 𝐴, 𝐵, 𝐶 undergo small variations.
To understand how 𝑍 varies, we use the linear approximation (1st-order Taylor expansion):
This expression tells us how much 𝑍 changes if each variable changes by a small 𝑑𝐴, 𝑑𝐵, 𝑑𝐶.
Now, if the small changes 𝑑𝐴, 𝑑𝐵, 𝑑𝐶 are due to experimental errors, we can rewrite them in terms of uncertainties:
where we added in quadrature because we assume the errors are independent.
If a quantity Z is obtained by raising a measurement A to a power n:
The relative error is multiplied by n:
Which is obtained by applying logarithmic differentiation:
For more complex functions that combine sums, products, and powers, the previous rules are applied to each operation in the uncertainty calculation. Given 𝑍 as a function of 𝐴, 𝐵, 𝐶, the absolute error is:
A tool to calculate errors on complex formulas was created by Ohio State University researcher Subhash Bose.
This formula derives from the propagation of uncertainty using partial derivatives, specifically from error propagation based on the Taylor expansion:
Squaring and summing, we obtain the expression for the total error.
This technique is valid when the errors are small and independent, because error propagation assumes a linear behavior around the measured value.
If the errors were large, the higher-order terms in the Taylor series expansion would become significant, making the approximation less accurate.
If the errors were correlated, then we would also have to include the covariances between the variables, complicating the propagation. For example, if A and B are not independent, the error on Z also depends on how A and B vary together. In that case, the formula becomes:
where ρ is the correlation coefficient between A and B.