SECU0006 Chap.4 Association and Statistical Evidence
Association and Statistical Evidence
Association and Statistical Evidence examines relationships without overstating what they prove. Scatterplots and correlation describe linear co-movement, while confounding, reverse direction, selection and measurement can alter interpretation. Evidence questions require comparing explanations and naming the limitation that constrains generalisation or causal language.
What this chapter covers
- 01
Scatterplots
- 02
Correlation
- 03
Nonlinear patterns
- 04
Confounding
- 05
Competing hypotheses
- 06
Causal boundaries
Worked example · free
Interpret a group difference
- 1Report the 0.7-point observed difference.
- 1Name the scale and group sizes.
- 1Avoid causal language without an identifying design.
Key terms
- Association
- A pattern in which values of variables vary together.
- Reverse causation
- A possibility that the proposed outcome influences the proposed cause.
- Selection bias
- Distortion arising when observed cases differ systematically from the target population.
Association and Statistical Evidence FAQ
Does zero correlation mean no relationship?
No. A curved pattern or opposing subgroup patterns can produce little overall linear correlation. Inspect a scatterplot and subgroup structure before concluding that the variables are unrelated.
Why compare competing explanations?
An observation can be rare under several hypotheses, so rarity under one does not alone establish another. The comparison must also account for dependence and the population to which the probabilities apply.
What is a useful limitation?
It identifies a specific feature such as sampling, measurement or timing and explains how it changes interpretation. A strong limitation names what evidence would reduce the uncertainty instead of ending with a generic caution.
Assessment move
For each relationship, draw the graph, state direction and strength, inspect outliers and subgroups, then list plausible confounders and reverse pathways. Replace causal verbs unless the design supports them. Practise explaining why a rare observation must be compared across hypotheses. Finish each answer with one concrete limitation and the additional evidence that would reduce it. Build relationship analysis in layers.
First identify the response, explanatory variable, population and time order. Plot raw observations and inspect curvature, clusters, missingness, influential points and subgroup patterns before calculating a coefficient. Next describe direction, form and strength without causal language. List plausible confounders, reverse pathways and selection mechanisms, then decide what the design can address.
Practise scenarios in which an overall association changes after grouping so that a single coefficient never substitutes for inspection. For evidence questions, write competing hypotheses and compare the same observation under each; do not convert a small probability under one hypothesis into certainty for another. When uncertainty measures are supplied, connect them to estimate precision and substantive magnitude.
When they are absent, report the observed pattern without inventing confidence. Use verbs deliberately: describes, differs and is associated with belong to observational evidence, while causes requires a defensible identification strategy. Keep an error log for correlation treated as causation, nonlinear patterns missed, rare-event fallacies, dependence ignored and limitations stated without consequences.
Close every written response with a limitation that changes interpretation and the additional sample, measure or design that would reduce it. This final move turns caution into a useful research recommendation. Strengthen interpretation by practising counter-cases. Start with a positive scatterplot and invent a plausible common cause that would produce both variables.
Then reverse the proposed direction and ask what temporal data would distinguish the stories. Create a curved relationship with near-zero linear correlation and a grouped relationship whose overall trend differs from subgroup trends. These examples make graphical inspection part of the method rather than decoration. For each scenario, separate descriptive magnitude from uncertainty and uncertainty from causality.
If a confidence interval is given, discuss its range in the measurement unit and whether substantively important values remain plausible. If a test result is given, do not treat a threshold label as an effect size. For evidence comparisons, write a table with hypotheses in columns and the observed evidence in rows, then reason about relative support.
State which numbers are illustrative and never transfer proprietary case details into practice questions. In assignment writing, connect every limitation to the direction in which it might alter the result when that direction is defensible; otherwise state the uncertainty it creates. Propose a realistic next study, such as repeated measurement, a clearer variable, a broader sample or an explicit comparison group.
The closing paragraph should distinguish what was observed, what remains uncertain and what action can reasonably follow now. During review, underline claims stronger than the design supports and identify the precise evidence that would justify each stronger version. Draft a new evidence claim and test whether its verb, population and limitation stay within the design actually described.
Working through Association and Statistical Evidence in SECU0006? Sia is AskSia’s AI Statistics tutor — ask any SECU0006 Association and Statistical Evidence question and get a clear, step-by-step explanation grounded in how SECU0006 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.