Unit 01 · Day 8
Patterns, Uncertainty & Causation
Move from seeing a pattern to judging what it can support. Separate signal from variation, distinguish random effects from systematic bias, test alternative explanations, and use causal language only when the study design earns it.
Learning objectives
- Describe a quantitative pattern with its direction, scope, variation, and anomalies rather than relying on vague trend language.
- Distinguish random variation from systematic bias and explain why they affect conclusions differently.
- Separate association from causation and generate plausible confounding, reverse-causation, selection, and coincidence explanations.
- Compare the causal strength of controlled experiments and observational studies without treating either as automatically perfect.
- Use uncertainty-aware language that matches the design and evidence.
- Identify additional evidence or a redesigned study that could discriminate among competing explanations.
Materials
- Science notebook
- Unit 01 Dataset Laboratory
- Day 06–07 helicopter data/graphs for comparison
- Khaemenes Scientific Calculator if arithmetic verification is useful
- Evidence Citation Studio when extending a claim with external sources
This lesson uses existing or fictional datasets and requires no hazardous materials. When extending the investigation into real outdoor observations, follow local supervision, weather, traffic, privacy, and site-safety requirements rather than assuming this classroom analysis authorizes fieldwork.
1. Phenomenon: a pattern appears
Imagine five shaded sites have lower surface temperatures than five nearby unshaded sites. The pattern may be scientifically interesting, but the graph alone does not tell us why it occurred.
Write (1) what was observed, (2) one possible explanation, and (3) one alternative explanation. Keep those three statements separate.
This separation is central to scientific reasoning: observation ≠ explanation ≠ proof of cause.
2. Describe patterns with scope
A useful pattern statement includes the variables, direction or shape, tested/observed range, variation, and notable exceptions.
Stronger observational statement: “Across the observed sites in this dataset, greater shade coverage was generally associated with lower measured surface temperature, although values varied among sites.”
The stronger version does not claim universality and does not silently convert an association into an experimental causal result.
3. Random variation and systematic bias
| Feature | Random variation | Systematic bias |
|---|---|---|
| Pattern | Unpredictable differences among measurements/observations | Consistent shift linked to instrument, method, sampling, or design |
| Example | Small stopwatch reaction differences among trials | Temperature sensor consistently reading high |
| Can repetition help? | Often helps characterize/average some variation | Not necessarily; repeated biased measurements can remain biased |
| Useful response | More repeated observations, better precision, characterize spread | Calibration/reference checks, redesign, better sampling/control |
4. Association has multiple possible explanations
When X and Y vary together, several structures are possible:
- X → Y: X contributes causally to Y.
- Y → X: the direction may be reversed.
- Z → X and Y: a third variable may influence both.
- Selection: the observations included may differ systematically from those not included.
- Coincidence/noise: an apparent pattern may arise in limited data without a stable underlying relationship.
Umbrellas are more common when pavement is wet. Give a third variable that explains both, then explain why the association does not establish that umbrellas wet pavement.
5. Confounding
A confounder is a variable related to both the explanatory condition and the outcome that can create or distort an apparent relationship. In the shade example, possible confounders might include surface material, time of day, moisture, wind exposure, building proximity, or site selection.
Choose two possible confounders. For each, explain how it could be associated with shade and independently affect surface temperature.
6. Experiment versus observational study
A well-designed controlled experiment deliberately manipulates an independent variable and uses controls/randomization where feasible to reduce alternative explanations. An observational study measures conditions as they occur. Observational studies are essential when manipulation is unethical, impossible, impractical, or would destroy the phenomenon being studied—but causal inference generally requires additional care.
| Question | Paper helicopter | Shade dataset |
|---|---|---|
| Was an independent variable deliberately changed? | Yes, if rotor length was manipulated | No, if existing sites were simply observed |
| Can conditions be controlled? | Many can be held reasonably constant | Many site differences may remain |
| Strongest immediate claim | Causal evidence within tested conditions if design was sound | Association within observed sites |
Neither label guarantees quality. A badly confounded experiment can be weaker than a carefully designed observational study.
7. Dataset Laboratory: shade & surface temperature
Analyze the fictional dataset without assuming the conclusion in advance.
- Identify the explanatory and response variables.
- Create or inspect an appropriate graph.
- Describe the direction, strength/consistency in plain language, range, and anomalies.
- State the strongest association claim supported by the observations.
- Generate at least three plausible alternative explanations/confounders.
- Choose the most important alternative and propose data or a study design that could test it.
- State what the dataset cannot establish.
8. From correlation to stronger causal evidence
No single checklist automatically proves cause, but causal confidence can increase when multiple forms of evidence converge—for example, a plausible mechanism, correct temporal order, controlled intervention, replication, dose/response patterns where scientifically meaningful, and alternative explanations that have been tested rather than ignored.
Design a safer, fairer investigation that could test whether shade itself contributes to surface-temperature differences. Specify what you would manipulate or match, what you would measure, and which confounder you are trying to reduce.
9. Uncertainty-aware scientific language
| Evidence situation | Appropriate phrasing |
|---|---|
| Observed numerical pattern | “The data show…” / “Within these observations…” |
| Observational relationship | “X was associated with Y…” |
| Controlled experimental result | “Changing X produced/was associated with a change in Y under these tested conditions…” |
| Proposed mechanism | “One explanation consistent with the evidence is…” |
| Limited evidence | “The current data do not distinguish between…” |
10. Evidence Citation Studio + Mentor causal-defense conference
If you extend the shade investigation with external scientific information, use Evidence Citation Studio to keep sourced evidence separate from your own dataset and interpretations. Cite sources; do not let a source substitute for analysis of the data in front of you.
Bring your causal map to the Science Mentor and defend:
- your strongest evidence-supported claim;
- the most serious alternative explanation;
- why the study design does or does not support causal language;
- what new evidence would discriminate between competing explanations;
- one sentence you deliberately weakened because the original wording overclaimed.
Mentor and ProTools support evidence reasoning; they do not determine the conclusion or replace the ≥80% mastery requirement.
11. Choose your pathway
| Pathway | Required reasoning |
|---|---|
| Foundation | Describe the dataset pattern, distinguish random/systematic effects, identify one confounder, and rewrite one causal overclaim as an association statement. |
| Core | Complete the dataset analysis, three alternative explanations, experiment/observation comparison, redesign challenge, and causal-defense conference. |
| Extended | Build a causal diagram with multiple variables; compare two plausible causal structures; identify what additional observation/intervention would distinguish them; evaluate whether selection bias could reverse or exaggerate the observed relationship. |
12. Exit check
- Why can repeated measurements reduce some random variation without correcting systematic bias?
- What is a confounder?
- Give two explanations for an association besides X causing Y.
- Why can an observational study be scientifically valuable even when it cannot by itself isolate cause?
- What makes causal language stronger in a controlled experiment?
- Rewrite “Shade causes lower temperature everywhere” into a statement justified by an observational dataset from a limited set of sites.
Save the pattern statement, random/systematic analysis, confounder map, Dataset Lab work, redesign proposal, and calibrated final claim. Day 08 completion records participation; Unit 01 mastery still requires ≥80% on designated mastery assessments and required constructed evidence.