Scientific Reasoning and Problem Solving - Variables, Data and Evidence
Scientific reasoning and problem solving is the transversal part of the ENP Alger access contest: it asks whether you can read a protocol, a table or a graph, identify what is being varied, and decide what the data actually support. The questions are deliberately plain in their arithmetic and demanding in their logic - the difficulty lies in distinguishing an observation from an inference and in spotting what an experiment failed to control.
What is tested
- Variables: independent (what the experimenter changes), dependent (what is measured), and controlled (held constant).
- Design: control group, replication, randomisation, and blinding where relevant.
- Data handling: mean, range or standard deviation, uncertainty, units, significant figures, graph shape, linearity and intercept.
- Errors: random error produces scatter and is reduced by averaging; systematic error produces a consistent bias and is not.
- Logic: correlation versus causation, necessary versus sufficient conditions, generalisation beyond the sample.
Worked example
A student measures the density of an aluminium block five times and obtains 2.81, 2.80, 2.82, 2.81 and 2.80 g/cm³; the accepted value is 2.70 g/cm³. The spread is about ±0.01, so the measurements are precise. But all five lie above 2.70, so the deviation is systematic, not random, and averaging more trials will not remove it. Typical causes are a zero error in the balance or a misread volume. The defensible conclusion is "precise but inaccurate - check the calibration", not "the accepted value is wrong".
Traps that cost marks
- Claiming a hypothesis is "proven" from a single experiment, or extending a conclusion beyond the population sampled.
- Reporting the range as if it were a standard deviation, or quoting an uncertainty with more digits than the measurements allow.
- Ignoring a confounder such as temperature, time of day or a change of operator between trials.
- Treating a y-intercept as physically meaningful when the model does not justify it.
- Calling a trend linear without checking the residuals or the units on both axes.
- Confusing precision (repeatability) with accuracy (closeness to the true value).