Does the Scientific Model Match the Evidence? A Short Student Routine
Students should state what the model predicts under specified conditions, compare that prediction with the corresponding observations, and examine both the size and pattern of any differences. Then they can judge whether the evidence supports using the model for that purpose, challenges a particular prediction, or leaves the decision unresolved. Agreement supports a model within the conditions checked; it does not establish that every part of the model is correct.
Start with a prediction students can check
Students can tell whether a scientific model matches the evidence by comparing what it predicts with what was observed under the same conditions. Ask them to identify the agreement, describe any mismatch, and decide what that comparison permits them to conclude.
A model can be a diagram, physical representation, equation, or simulation. Students need to use its relationships to say what should happen: “If this model applies under these conditions, we should observe ___.” The National Research Council describes model evaluation as an iterative comparison between predictions and the real world. [Framework, Chapter 3](https://www.nationalacademies.org/read/13165/chapter/7)
For New York science classrooms in grades 8–12, the routine below offers one practical way to organize that comparison. It is an editorial synthesis, not a tested intervention or an official Regents scoring rubric.
Use four steps: predict, pair, compare, decide
1. Predict. Write one observable consequence of the model, including the conditions. Replace “the model shows motion” with “the model predicts that the cart travels equal distances during equal time intervals.” For a diagram, explain which arrow or relationship produces the prediction. If students cannot derive an observable consequence, clarify the model before judging its fit.
2. Pair. Put each prediction beside the corresponding observation. Match the variable, units, time, location, and relevant starting conditions. A prediction for position at three seconds needs the position observed at three seconds. Keep predictions and observations in separate columns so students can inspect their relationship.
3. Compare. Describe where they agree and where they differ. For numerical predictions, calculate observed minus predicted and retain the units. Look across the comparisons: Are differences small, consistently positive, growing over time, or concentrated under one condition? For qualitative predictions, compare the predicted direction, order, presence, or absence with the record.
4. Decide. Choose a conclusion with a boundary: “supported for this purpose under these conditions,” “challenged by this mismatch,” or “not yet resolved.” Name the particular prediction and evidence behind the decision. Then identify whether to retain the model provisionally, revise a relationship or assumption, or obtain a more informative observation.
NGSS Appendix F supports the underlying practice: students develop and revise models using evidence, make predictions, and evaluate models’ merits and limitations. The four-step sequence is a suggested classroom implementation of those ideas. [NGSS Appendix F, page 19](https://www.nextgenscience.org/sites/default/files/resource/files/Appendix%20F%20%20Science%20and%20Engineering%20Practices%20in%20the%20NGSS%20-%20FINAL%20060513.pdf)
Work through a cart example
This is an invented instructional example, not a report of an experiment. Model A represents a cart moving at a constant speed of 20 centimeters per second from a starting position of zero. At one, two, and three seconds, it predicts positions of 20, 40, and 60 centimeters.
Supply this fictional observation record: at one second, 20 centimeters; at two seconds, 45 centimeters; at three seconds, 75 centimeters. For this exercise only, specify that position uncertainty is ±1 centimeter, timing uncertainty is negligible, and the model’s predictions are treated as exact.
Pairing the values gives observed-minus-predicted differences of 0, +5, and +15 centimeters. The first prediction agrees with the observation. The later observations lie above the predictions by amounts larger than the stated position uncertainty.
The pattern matters: the model increasingly underpredicts position. A student conclusion could be: “Model A matches the first position but does not adequately represent the full one-to-three-second record under the stated uncertainty. The growing positive differences challenge its constant-speed relationship.”
A useful next move is to propose a model in which speed changes, then test its predictions against additional observations. These three positions alone do not identify what caused the motion. Keep the original record when revising; changing the observations to fit the model would erase the evidence being evaluated.
Decide whether a difference matters
Before inspecting the outcome, establish what the comparison is intended to check. A model intended to predict only whether a quantity increases should be evaluated first against that direction. A model intended to predict numerical values requires a numerical comparison. Matching the direction can support one prediction while leaving numerical accuracy untested.
For quantitative work, consider the uncertainty of both measurements and predictions when that information is available. Avoid a universal rule such as “within 10 percent means correct.” If uncertainty information is missing, students can report the differences and identify what additional information they need to judge whether those differences matter for the intended use.
Check the whole record rather than selecting the closest point. In statistical model fitting, NIST defines residuals as observed values minus corresponding predictions and explains why their patterns help reveal poor fit. Its guidance also cautions that a high R-squared value alone does not establish a good fit. The classroom comparison here borrows that diagnostic idea without requiring a formal statistical analysis. [NIST: How can I tell if a model fits my data?](https://itl.nist.gov/div898/handbook/pmd/section4/pmd44.htm)
When a mismatch appears, check transcription, units, calculations, measurement procedures, and whether the observed conditions match the model’s assumptions. Treat each as a possibility requiring evidence. “Human error” does not explain a discrepancy unless students identify a specific problem and show how it could affect the comparison.
Handle agreement and inconclusive evidence carefully
Agreement supports the prediction checked under the conditions examined. Keep the conclusion within that range. The Framework emphasizes that models contain approximations and assumptions that limit their application and predictive precision. [Framework, Chapter 3](https://www.nationalacademies.org/read/13165/chapter/7)
Two models may make the same prediction for the available observations. In that case, the evidence does not distinguish them. Ask students to find a condition under which the predictions differ and identify an observation that could resolve the choice.
Use a brief closing statement: “The model predicts ___. The record shows ___. The agreement or mismatch is ___. Therefore, I would ___ for these conditions. The next useful observation is ___.” Accept an unresolved judgment when students explain exactly what evidence is missing.
If students need help identifying the recorded evidence, use the [observation-or-inference routine](/blog/observation-or-inference-science-data-routine). If they have completed the comparison but need to improve their written justification, use the guide to [actionable feedback on science explanations](/blog/actionable-feedback-science-explanations).