Natural Sciences: Plot my own data

What you'll be able to do: Work with the numbers your own practical actually produced, rather than the idealised curve the textbook prints, which is where the science is.

Why your messy data is better than clean data

Textbook graphs are smooth. Real ones aren't, and students conclude their practical went badly.

It usually didn't. The deviation is the content: it tells you about measurement error, about the conditions the model assumes, about where the idealisation stops holding. A student who plots their own scattered points and reconciles them against the expected relationship is doing something a student who copies the textbook curve isn't doing at all.

And there's an advantage unique to your own data: you know what the numbers mean and how they were collected. A nonsensical result is obviously nonsensical rather than merely unexpected, and you know which point was the one where the stopwatch slipped.

The loop

1. Predict the shape before plotting, linear, exponential, plateauing, sigmoid, and say why. (Article 03.)
2. Plot it yourself, choosing the axes and saying which variable you think explains which.
3. Ask what the chart is hiding. Every representation suppresses something.
4. Re-plot differently: log axis, different scale, residuals, and see whether your conclusion survives.
5. Reconcile with the expected relationship. Where does yours deviate, and is that measurement error, a condition the model assumes, or a real effect?
6. Ask what you'd need to claim causation. Almost always: more than you have.

Here's my practical data [paste] and what I expected: [prediction]. Plot it, Where does it deviate from the expected relationship, and for each deviation: is that most likely measurement error, an assumption of the model being violated, or something real?

That three-way question is the one that turns a write-up's "errors and improvements" section from a ritual into an analysis.

Across the sciences

Chemistry: rates. Your concentration-time data won't be a clean curve. Ask whether the deviation is at the start (mixing time), the end (reagent exhaustion), or throughout (systematic).

Chemistry: titrations. Plot the whole curve rather than recording the endpoint. The shape around the equivalence point is the chemistry, and recording one number discards it.

Physics: motion. Plot your own timings. Air resistance shows up as systematic deviation at higher speeds, which is a much better introduction to "the model assumes no drag" than a sentence.

Physics: springs. Hooke's law plotted from real data has a proportional region and then doesn't. Finding your own elastic limit is more convincing than reading about it.

Biology: enzymes. Rate against temperature from your own results, showing the rise and the fall. The denaturation side arrives as data rather than as a claim.

Biology: populations or growth. Real growth data on a log axis is a transformative moment for understanding exponential processes.

Earth science: field measurements. Grain size against distance, temperature against depth. Spatial data plotted yourself makes the gradient real.

Astronomy: observations. Your own brightness or position measurements over nights. Small datasets and genuine uncertainty.

The thing to be honest about

Twenty points from one practical is enough to learn the skill and not enough to establish anything. Say so in the write-up rather than hedging. A student who writes "this is consistent with the expected relationship, though with n=8 and no repeats I cannot distinguish this from several alternatives" is demonstrating more understanding than one who writes "this proves".

Pitfalls

1. Not predicting the shape first. You'll look at the plot and feel you expected it.
2. Plotting the textbook curve instead. Discards the entire exercise.
3. One plot only. A conclusion that survives a single representation isn't a conclusion about the data.
4. Treating deviation as failure. It's the informative part.
5. Causal language from correlational data, which is more tempting with your own results because you have a theory about them.
6. The tell: your graph is neat and you can't say what you'd expected or where it differs.

Try this today

Take the data from your most recent practical, including the point you were tempted to discard.

Predict the shape before plotting. Then plot it, and ask the three-way question about each deviation: measurement error, violated assumption, or something real?