Data Evaluation and Drawing Conclusions

From measurements to evidence

Collecting readings is only the beginning of an investigation. Before using them to reach a conclusion, you must ask how trustworthy they are, whether they show a convincing pattern and whether that pattern answers the question being investigated. No measurement is perfectly certain because instruments have limited resolution and their users may introduce variation.

It helps to separate four questions: What do the data show? How good are the data? What conclusion follows? How does that conclusion affect the hypothesis? A neat-looking set of numbers is not automatically strong evidence.

Accuracy and precision answer different questions

Accuracy is closeness to the true or accepted value. Suppose a standard mass is known to be 100.0 g. Measurements centred near 100.0 g are more accurate than measurements centred near 105.0 g. Accuracy can only be assessed directly when there is a trustworthy value or standard for comparison.

Precision concerns the spread of repeated measurements. Readings of 104.9 g, 105.0 g and 105.1 g are precise because they cluster closely, but they are inaccurate if the accepted value is 100.0 g. Conversely, readings scattered on either side of 100.0 g might have an accurate mean while having poor precision.

Instrument resolution is the smallest change that an instrument can distinguish. Finer resolution can reduce reading uncertainty and improve precision, but extra decimal places do not guarantee accuracy. A badly calibrated instrument can produce tightly grouped readings that are all wrong.

Repeatability and reproducibility

Results are repeatable when the same investigator obtains similar results using the same method and equipment under the same conditions. Repeating readings during one investigation therefore provides evidence about repeatability.

Results are reproducible when similar results are obtained by different investigators using different equipment. Reproducing a result provides evidence that it does not depend only on one person or one apparatus arrangement. Comparing another suitable method can also help reveal weaknesses that both sets of repeats might otherwise hide.

Neither repeatability nor reproducibility guarantees accuracy. A faulty instrument can give the same wrong reading repeatedly, while different investigators may share a method containing the same systematic bias.

Random error, systematic error and uncertainty

A random error varies unpredictably between readings, producing scatter. Examples include variation in reaction time, small environmental fluctuations and estimating a pointer position between scale marks. Taking repeat measurements shows the spread of the data. Calculating a mean reduces the influence of readings that happen to be unusually high or low, although it does not remove measurement uncertainty.

A systematic error moves results consistently away from the true value. For example, a balance that reads 2 g when empty may make every measured mass about 2 g too high. Repeating and averaging these readings will not remove the offset. The cause must instead be addressed, perhaps by zeroing or calibrating the balance or changing a flawed procedure.

Uncertainty expresses the limited confidence available in a measured value. It is not evidence that the experiment has failed; it is an honest description of the limits of the measurement. Large scatter or wide uncertainty intervals makes small apparent differences less convincing.

Treating an outlier scientifically

An outlier lies noticeably away from the pattern formed by the other data. It should initially be treated as data rather than deleted to make a pattern look stronger.

Check the original record, units, calculation, apparatus and experimental conditions. Repeat that measurement if possible. Rejecting it is justified when there is a specific reason, such as a recording mistake, an instrument malfunction or a known failure to follow the method. If no cause is found, retain it and discuss how it affects the conclusion. An unexpected result might represent genuine variation or reveal an uncontrolled factor.

Evaluating an experimental strategy

An evaluation should identify a particular weakness and connect an improvement to the quality of the data:

• Repeating each measurement and calculating a mean reduces the influence of random variation and allows repeatability and spread to be assessed.
• Taking more readings across a suitable range makes a pattern easier to judge and reduces the influence of any single point.
• Zeroing or calibrating an instrument against a known standard addresses possible systematic error and can improve accuracy.
• Selecting equipment with suitable range and finer resolution can reduce reading uncertainty and improve precision.
• Replacing a hand-operated stopwatch with a light gate or electronic sensor can reduce random variation caused by reaction time.
• Keeping relevant control variables constant makes it more reasonable to connect changes in the dependent variable with the independent variable.
• Giving a precise, standardised method allows another investigator to repeat the work and provides a fairer test of reproducibility.

The explanation matters. “Use better equipment” is too vague: name the relevant feature and explain which problem it addresses. Also recognise limits. Repeats can reduce the effect of random error but cannot correct a systematic offset.

A useful further investigation extends or checks the evidence. It might explore a wider range, take extra readings around an unexpected region, repeat the work with different equipment or test whether the result is reproducible. It should address a question raised by the original investigation rather than merely repeat it without a purpose.

Turning observations into a reasoned conclusion

Data may be numerical, graphical, symbolic or diagrammatic. A sound interpretation follows a clear chain:

1. Identify the relevant pattern, difference or absence of a pattern.
2. Support that description with values or other specific features of the data.
3. Consider scatter, uncertainty and outliers.
4. Make an inference that answers the investigation question.
5. Explain the scientific reasoning using appropriate terminology.

Describe only what the evidence warrants. If a graph generally rises, the data support the statement that one measured quantity increases as the other increases over the tested range. They do not justify pretending that every point follows the pattern exactly or extending the conclusion far beyond the investigated range.

A conclusion should therefore combine a claim, relevant evidence and scientific reasoning. For example, you might state that the outcome increased across most of the tested range, support this with values, note that repeats were closely grouped and then conclude that the predicted relationship is supported within that range. Any anomaly, large uncertainty or other important limitation should qualify the strength of the conclusion.

Evidence changes confidence in a hypothesis

A hypothesis is a tentative, testable explanation. It is used to make a prediction, which is then compared with observations or experimental results.

Agreement between good-quality data and the prediction increases confidence in the hypothesis. It does not prove the hypothesis: another explanation might fit the same results, and later tests might produce different evidence. Confidence becomes stronger when accurate, precise and reproducible evidence supports predictions across varied conditions.

Disagreement between a prediction and trustworthy data decreases confidence. The hypothesis, the prediction drawn from it or both may need to be changed or rejected. However, the quality of the investigation must first be considered. A mismatch produced by an uncontrolled variable, systematic error or highly uncertain measurements is weaker evidence against a hypothesis than a reproducible mismatch from a well-designed investigation.

The final judgement should be proportional to the evidence: state whether confidence increases or decreases, cite the relevant data, evaluate their quality and acknowledge limitations. Scientific explanations remain open to further testing and revision.

Data quality

• Accurate: close to the true or accepted value.
• Precise: repeated values have little spread.
• Repeatable: similar results from the same investigator, method and equipment.
• Reproducible: similar results from different investigators using different equipment.

Precision does not guarantee accuracy.

Errors, uncertainty and outliers

• Random error causes unpredictable scatter; repeats and a mean reduce its influence.
• Systematic error shifts readings consistently; repeating does not remove it.
• Every measurement has uncertainty.
• Retain an outlier unless investigation reveals a defensible reason to reject it.

Evaluating a strategy

Use the chain:

weakness → specific improvement → effect on data quality

Examples of valid links include calibration to improve accuracy, finer-resolution equipment to improve precision, repeats to assess repeatability and an independent investigator using different equipment to test reproducibility. Further work could extend the range, investigate an anomaly or check the finding using another arrangement.

Drawing a conclusion

State the pattern, cite relevant data, consider scatter, uncertainty and anomalies, and then make the scientific inference. Limit the conclusion to the range and conditions actually investigated.

Judging a hypothesis

Agreement with a prediction increases confidence; trustworthy disagreement decreases it. Evidence can support a hypothesis but cannot prove it. Judge the strength of support using the quality and amount of the evidence.