Test a model or prediction against ACT Science data: decide whether results support it, which data would strengthen or weaken it, and how to refine a model that partly fails.
Compare what the model PREDICTS to what the data OBSERVED at each input.
Match = support; divergence = the model needs refining or rejecting.
Classify a model as well supported, partially supported, or unsupported.
The qualified "supported only up to X" answer is often the accurate one.
Confirm predictions in an untested range, or beat a rival model.
Weaken it by finding a clear prediction miss or a simpler equal fit.
Change the model only where the data broke it; keep what already works.
Reject refinements that contradict the range where the model succeeded.
The better model has smaller, more consistent gaps from the data.
Close-everywhere beats nails-one-point-misses-the-rest.
A model on ACT Science is any proposed rule, equation, mechanism, or prediction that claims to explain or forecast data. It might be a formula ("output doubles each time the input rises by 10"), a proposed mechanism ("the gas escapes through tiny pores"), or a specific numerical prediction. Evaluating a model means one thing: compare what the model predicts to what the data actually shows. Where they match, the model is supported; where they part ways, the model needs refining or rejecting.
Start every model question by writing down two things: what the model predicts and what the data observed. Suppose a model says a spring's stretch is directly proportional to the hanging mass — double the mass, double the stretch.
| Mass (g) | Predicted stretch (cm) | Observed stretch (cm) |
|---|---|---|
| 100 | 2.0 | 2.0 |
| 200 | 4.0 | 4.1 |
| 300 | 6.0 | 5.9 |
| 400 | 8.0 | 7.2 |
For 100–300 g the observed values hug the prediction, so the model works there. At 400 g the prediction (8.0) badly overshoots the observation (7.2): the model breaks down at high mass, likely because the spring is nearing its limit. A good evaluation names both the range where the model succeeds and the point where it fails — the ACT rewards that precision over a flat "the model is right" or "the model is wrong."
Models are rarely all-or-nothing. Train yourself to place a model in one of three buckets:
When choices offer "the model is supported," "supported only up to 300 g," and "not supported," the middle, qualified statement is often the accurate one. Beware answers that overstate the model's success or dismiss it entirely.
To strengthen a model, you either confirm its prediction where it has not yet been tested or rule out a rival explanation.
To weaken a model, do the opposite: find a case where its prediction clearly misses, or show a simpler model explains the same data equally well. On the spring example, a trial at 500 g whose observed stretch again falls far below the prediction would further weaken the "always proportional" model.
The subtlest model questions ask how to fix a model that partly fails. The rule: change the model only as much as the data demands, keeping what already works.
For the spring, the proportional model works up to about 300 g and then over-predicts. A sensible refinement is: "stretch is proportional to mass for light loads but increases more slowly once the spring nears its stretching limit." That keeps the successful low-mass behavior and adds a qualifier exactly where the data broke the original rule. A refinement that throws out the whole model, or that contradicts the data where the model did work, is wrong.
Sometimes two models compete to explain one data set, much like Conflicting Viewpoints but with numbers. The better model is the one whose predictions sit closer to the observed values across more of the range. Compute or eyeball the gap between each model's prediction and the data at several points; the model with the smaller, more consistent gaps wins. A model that nails one point but misses the rest loses to one that is close everywhere.
Treat a model as a promise about the numbers, then hold it to the data. The ACT is not asking whether the model is elegant — only whether the measurements kept its promise.
A model predicts a car's fuel use rises by exactly 1 L for every 10 km/h of extra speed. Table 1 compares the prediction to measurements.
| Speed (km/h) | Predicted fuel (L per 100 km) | Observed fuel (L per 100 km) |
|---|---|---|
| 60 | 6.0 | 6.0 |
| 70 | 7.0 | 7.1 |
| 80 | 8.0 | 7.9 |
Do the data support the model across the tested range?
A model says a plant's growth rate doubles for every extra hour of light. Table 1 lists prediction vs. observation.
| Light (h) | Predicted growth (mm/day) | Observed growth (mm/day) |
|---|---|---|
| 4 | 2 | 2 |
| 5 | 4 | 4 |
| 6 | 8 | 8 |
| 7 | 16 | 9 |
At what point does the model stop matching the data?
Using the light/growth data above (matches through 6 h, misses badly at 7 h), how should the "growth doubles each hour" model be described?
A model predicts that a chemical's reaction rate rises by 3 units for each 10 °C. It has been confirmed from 20 °C to 50 °C. A chemist wants to strengthen confidence that the model is generally valid. Which new test would best do so?
The "growth doubles each hour of light" model matched through 6 hours but predicted 16 mm/day at 7 hours while the plant grew only 9. How should the model be refined?
Two models predict a cooling cup's temperature. Table 1 shows both predictions against the measured values.
| Time (min) | Observed (°C) | Model X (°C) | Model Y (°C) |
|---|---|---|---|
| 0 | 90 | 90 | 90 |
| 10 | 66 | 70 | 65 |
| 20 | 50 | 55 | 51 |
| 30 | 40 | 44 | 39 |
Which model better fits the data, and why?
Declaring a model simply "right" or "wrong" when it actually fits part of the range and fails elsewhere.
Compare prediction to observation at every input and name the boundary. The accurate answer is often "supported only up to X."
Thinking that repeating a test in the range where a model already works strengthens it.
Strengthen a model by confirming it in an UNtested range, or by showing it beats a rival. Re-confirming the known range adds little.
Refining a model by throwing the whole thing out, even the part that matched the data.
Keep what worked and adjust only where the data forced a change. A refinement that contradicts the successful range is wrong.
Judging two competing models by a single data point where one happens to be exact.
Compare the gaps across the whole range. A model that is close everywhere beats one that nails one point and misses the rest.
A model predicts that a balloon's volume rises by 5 mL for each 1 °C of warming. Table 1 compares prediction and measurement.
| Temp (°C) | Predicted volume (mL) | Observed volume (mL) |
|---|---|---|
| 20 | 500 | 500 |
| 25 | 525 | 524 |
| 30 | 550 | 551 |
Do the data in Table 1 support the model across the tested range?
A model claims a bacteria colony's count triples every hour. Table 1 lists the prediction and the observed count.
| Hour | Predicted count | Observed count |
|---|---|---|
| 1 | 300 | 300 |
| 2 | 900 | 890 |
| 3 | 2,700 | 2,650 |
| 4 | 8,100 | 4,000 |
At which hour does the model first clearly fail to match the data?
A model predicts a metal beam sags in proportion to the load on it. Measurements match the prediction for loads of 100–400 kg but the beam sags far more than predicted at 500 kg and 600 kg. How is the model best described?
A model predicting a river's flow from rainfall has been confirmed for rainfalls of 10–40 mm. A hydrologist wants to strengthen the case that the model is broadly valid. Which new test would best strengthen it?
A model that a spring's stretch is proportional to load matched loads up to 300 g but over-predicted the stretch at 400 g and 500 g, where the spring neared its limit. Which refinement is best supported by these results?
Two models predict how far light passes into a liquid. Table 1 shows both predictions against the measured depth.
| Dye (mg/L) | Observed depth (cm) | Model P (cm) | Model Q (cm) |
|---|---|---|---|
| 2 | 80 | 79 | 72 |
| 4 | 55 | 56 | 60 |
| 6 | 38 | 37 | 30 |
| 8 | 26 | 27 | 34 |
Which model fits the data better, and on what basis?
Judge whether ACT Science data actually supports a stated conclusion, tell a supported claim from an overreach, and name the new evidence that would confirm a result.
Master the ACT Science Conflicting Viewpoints passage: compare two scientists' hypotheses, find where they agree and disagree, and match new evidence to the side it supports.