AMU MATH302 Week 7 Study Guide: Correlation and Regression

Use your current classroom sequence

Exact weekly order and assessment labels can differ by section or term. Use the syllabus in your current classroom as the controlling source. The numbered route addresses a common search pattern; its concept map is an editorial study sequence derived from current official scope and a clearly historical archived example, not a claim about your current module.

Topic map and distinctions

Correlation describes the direction and strength of a linear association between two quantitative variables. It is unitless, bounded from −1 to 1, sensitive to outliers, and does not establish causation. A regression line predicts the response from an explanatory variable. The slope is the predicted response change for a one-unit increase in the explanatory variable; the intercept is the predicted response when the explanatory value is zero, which may or may not be meaningful in context.

The coefficient of determination describes the share of observed response variation accounted for by the fitted linear relationship in the analyzed data. A residual is observed response minus predicted response. Residual plots help reveal nonlinearity, changing spread, or unusual cases that a single fit statistic can hide. Prediction within the observed explanatory range is interpolation; prediction beyond it is extrapolation and requires caution. Excel output must be mapped to variables, units, and the original question before a conclusion is written.

Concept diagnostics and deeper checks

Always begin with the scatterplot because correlation and regression summaries can hide form. Look for direction, approximate linearity, clusters, gaps, unusual x values, unusual y values, and influential cases. Correlation measures only linear association and is unchanged by unit conversions that preserve direction. It can be near zero when a strong curved relationship exists, so the plot remains essential.

Regression assigns different roles to variables. The explanatory variable supplies x and the response supplies y. Swapping them changes the fitted equation because prediction errors are measured vertically in the usual least-squares setup. Interpret slope with response units per explanatory unit. Interpret the intercept only if x=0 is meaningful and supported. R-squared summarizes fit for the analyzed data, while residuals show what the line missed case by case.

Prediction needs a domain check. Interpolation uses x values inside the evidence range, though uncertainty still remains. Extrapolation assumes the relationship continues where no observations support it. A narrow-looking fitted line does not erase response variation, and a high correlation does not identify a mechanism. When a potential confounder could influence both variables, state that causal attribution is not established.

Method and formula selection

The Correlation-and-Regression Interpretation Grid asks three questions before calculation: what is the target, what data structure is present, and what assumption changes the method? Use the decision rows below as a compact selection table. Write the method choice in words before entering values so a function or familiar formula cannot conceal a mismatch.

Fully original fictional worked example

A fictional food-rescue program records route distance x in kilometers and delivery duration y in minutes for 24 routes spanning 3 to 18 kilometers. A fitted line is predicted duration = 11.2 + 2.6(distance), with correlation 0.82 and R² = 0.67. The positive correlation indicates a fairly strong positive linear association in these routes.

The slope predicts 2.6 additional minutes for each one-kilometer increase in route distance, on average within the observed range. The intercept predicts 11.2 minutes at zero kilometers, but zero lies outside the observed range and may not represent a meaningful delivery. R² says about 67% of the observed variation in duration is accounted for by its linear relationship with distance in this sample; it does not say that distance causes exactly 67% of duration.

For a 10-kilometer route, the fitted value is 37.2 minutes. If the observed duration was 41 minutes, the residual is 3.8 minutes: observed minus predicted. Predicting a 40-kilometer route would be extrapolation and is not supported merely because the equation can calculate it.

Excel check without hiding the reasoning

Before reading output, label response y and explanatory x. Verify the slope sign against the scatterplot and compare the displayed R² with the squared correlation in simple linear regression when appropriate. Recompute one fitted value and residual manually. Inspect residuals for pattern rather than assuming a high R² guarantees fit. Retain units in coefficient interpretations and avoid copying raw coefficient tables as a conclusion.

Original practice questions and reasoning checks

1. Correlation is −0.76. Describe direction and linear strength cautiously. Reasoning check: It indicates a fairly strong negative linear association, subject to the scatterplot and outlier check.

2. A slope is 4.2 dollars per hour. Interpret it. Reasoning check: Predicted response increases by 4.2 dollars for each additional explanatory hour within the fitted context.

3. Observed y is 31 and predicted y is 27. Find the residual. Reasoning check: 31−27=4, so the model underpredicted by 4 response units.

4. The observed x range is 5–30. Classify prediction at x=55. Reasoning check: Extrapolation; the fitted relationship may not persist that far.

5. Can R² = 0.90 establish causation? Reasoning check: No. Design, confounding, direction, and model validity still control causal claims.

Common wrong-answer patterns

Common errors reverse explanatory and response variables, interpret slope without units, force a meaningful intercept at an impossible zero, call correlation a causal effect, describe R² as the percent of cases predicted correctly, calculate residual as predicted minus observed, ignore influential outliers and curvature, and extrapolate because the formula returns a number.

How to check your own answer

Inspect the scatterplot first. Name x and y with units. Check correlation and slope signs for agreement. Substitute one observed x into the equation and calculate its residual as observed minus predicted. Read R² together with residual pattern. Mark the observed x range before predicting. State association, not causation, unless the design independently supports a causal interpretation.

Turn one solved example into reusable skill

Using the Correlation-and-Regression Interpretation Grid, finish the fictional example, close the calculation, and reconstruct its decision path from memory. Write the target, data structure, method, assumptions, key substitution, output, and interpretation on separate lines. Then change one condition—such as dependence, sample size, variable type, tail direction, pairing, or distribution shape—and explain whether the same method survives. This contrast practice is more durable than memorizing the displayed numbers.

Create a two-column error log. In the first column, record the earliest decision that failed: translation, classification, model, formula, software input, arithmetic, or interpretation. In the second, write a future check that would catch it. Use the Correlation-and-Regression Interpretation Grid as the organizing label, but express the check in your own words. Rework only original practice or material you are authorized to use; never build the log from uploaded restricted assessments.

Teach the Correlation-and-Regression Interpretation Grid result aloud without looking at the page. A complete explanation names why the method fits, what the output means in context, and one claim the evidence cannot support. If you can compute but cannot explain those three parts, return to the topic map. If you can explain but cannot reproduce the arithmetic, return to the transparent setup and Excel audit. Mastery requires both paths to agree.

Build a miniature formula card only after the reasoning is stable. Put the trigger question above the relationship, define every symbol with units, list the assumptions beside it, and place one reasonableness check below it. On the reverse, write a situation where the relationship should not be used. Attach the card to the Correlation-and-Regression Interpretation Grid rather than to a copied prompt. During review, cover the formula and recover it from the decision structure. This tests understanding while reducing the risk that a familiar-looking question activates the wrong procedure.

Short test-preparation checklist

Before beginning: identify the concept family and rewrite the target in your own words. During work: label inputs, units, distribution or parameter, and assumptions; keep enough precision to reproduce the result. Before finishing: use the Correlation-and-Regression Interpretation Grid, perform an independent numerical or graphical check, and read the interpretation for scope and overclaiming. If current instructions use different notation or software, follow those instructions while preserving the same reasoning trail.

Related MATH302 resources

• MATH302 Statistics Help for AMU Students
• AMU MATH302 Week 6 Study Guide: Hypothesis Testing
• AMU MATH302 Week 8 Study Guide: Statistical Analysis and Interpretation
• AMU MATH302 Final Exam Study Guide: Method Selection, Excel, and Answer Checks