Comprehensive Study Guide to Scatter Plots, Linear Regression, and Inferential Confidence Intervals

Scatter Plot Analysis and Correlation

• Components of Bivariate Association:

  • When describing a scatter plot of bivariate numerical data, four primary characteristics must be evaluated: shape (form), direction, strength, and the presence of any outliers or unusual features.

  • The axes of the scatter plot range from $20$ to $85$ on both the horizontal ($X$) and vertical ($Y$) dimensions.

• Statistical Description of the Distribution:

  • Shape: Linear. The data cloud follows a generally straight-line path across the coordinate plane rather than demonstrating curved or non-linear curvature.

  • Direction: Negative. As values on the horizontal axis increase from $20$ to $85$, values on the vertical axis systematically decrease from $80$ down to $20$.

  • Strength: Moderate. The points are moderately scattered around the underlying linear trend line rather than being extremely tightly clustered or widely dispersed.

  • Correlation Coefficient ($R$): The estimated Pearson correlation coefficient is approximately $R \backsim -0.6$.

  • Outliers: No outliers. There are no isolated observations that diverge substantially from the general negative linear trajectory.

Confidence Intervals for Population Means (One-Sample t-Interval)

• Problem Framework and Context:

  • A cheese shop advertises and sells blocks of cheese labeled as one pound ($16\,\text{oz}$).

  • An investigation is conducted to estimate the true average mass ($\mu$) of the cheese blocks to test whether the store's stated claim of $16\,\text{oz}$ is accurate.

  • Sample summary statistics collected from the shop:

  • Sample size: $n = 40$

  • Sample mean: $\bar{x} = 15.29\,\text{oz}$

  • Sample standard deviation: $s = 1.69\,\text{oz}$

  • Target confidence level: $95\%$

• The FRED Procedure for Statistical Inference:

  • The FRED acronym defines a four-step framework for inferential procedures:

  • F (Formula / Framework): Identify the parameter of interest and select the appropriate sampling distribution and inferential formula.

  • R (Requirements / Conditions): Verify all technical assumptions necessary for valid inference.

  • E (Execution / Estimation): Calculate the point estimate, standard error, margin of error, and the final confidence interval boundaries.

  • D (Decision / Determination): Provide a formal contextual interpretation of the interval.

• Step R: Requirements:

  • Random Sample Condition: A random sample of $n = 40$ blocks of cheese was selected, ensuring independence among observations.

  • Nearly Normal Condition: The underlying distribution of sample values is symmetric with no notable outliers. Because the sample size $n = 40 \ge 30$, the Central Limit Theorem further guarantees that the sampling distribution of the sample mean ($\bar{x}$) is approximately normal.

  • Condition Status: Both conditions are fully met.

• Step E: Execution:

  • Degrees of Freedom ($df$):

  • $df = n - 1 = 40 - 1 = 39$

  • Critical Value:

  • For a two-tailed $95\%$ confidence interval with $df = 39$, the critical $t$-value is $t^* \approx 2.023$.

  • Standard Error ($SE$):

  • $SE = \frac{s}{\sqrt{n}} = \frac{1.69}{\sqrt{40}} \approx 0.2672\,\text{oz}$

  • Margin of Error ($ME$):

  • $ME = t^* \times \frac{s}{\sqrt{n}} \approx 2.023 \times 0.2672 \approx 0.54\,\text{oz}$

  • Calculated Confidence Interval Boundaries:

  • $(14.75, 15.83)\,\text{oz}$

  • Substantive Meaning: Because the entire $95\%$ confidence interval lies strictly below $16\,\text{oz}$, the empirical data suggest that the true average block mass is less than the advertised one-pound ($16\,\text{oz}$) claim.

Confidence Intervals for Population Proportions (One-Sample z-Interval)

• Problem Framework and Context:

  • The cheese shop sells both local and imported cheeses and seeks to estimate the true population proportion ($p$) of total cheese sales accounted for by imported varieties.

  • Sample parameters collected:

  • Sample size: $n = 250$

  • Sample proportion of imported cheese: $\hat{p} = 18\% = 0.18$

  • Observed count of successes: $x = n \times \hat{p} = 250 \times 0.18 = 45$

  • Observed count of failures: $n - x = 250 - 45 = 205$

  • Confidence level: $95\%$

• Underlying Conditions (Full FRED Reference):

  • Randomness: The $250$ blocks sold represent an appropriate sample of overall sales.

  • Success/Failure Condition:

  • $n\hat{p} = 250 \times 0.18 = 45 \ge 10$

  • $n(1 - \hat{p}) = 250 \times (1 - 0.18) = 205 \ge 10$

  • Both expected counts are greater than or equal to $10$, satisfying the condition for normal approximation.

• Step E: Execution:

  • Critical Value ($z^*$):

  • For a $95\%$ confidence level, $z^* = 1.96$.

  • Standard Error of the Proportion ($SE$):

  • $SE = \sqrt{\frac{\hat{p}(1 - \hat{p})}{n}} = \sqrt{\frac{0.18 \times 0.82}{250}} = \sqrt{\frac{0.1476}{250}} = \sqrt{0.0005904} \approx 0.024298$

  • Margin of Error ($ME$):

  • $ME = z^* \times SE = 1.96 \times 0.024298 \approx 0.04762$

  • Interval Computation:

  • Lower limit: $0.18 - 0.04762 = 0.13238$

  • Upper limit: $0.18 + 0.04762 = 0.22762$

  • Resulting interval: $(0.13238, 0.22762)$

• Step D: Decision and Contextual Interpretation:

  • With $95\%$ confidence, the true proportion of cheese that is imported is between $13.24\%$ and $22.76\%$.

Bivariate Data and Least Squares Regression Analysis

• Problem Setup and Observed Sales Data:

  • The cheese shop sells complementary retail items, including beef jerky and jam.

  • The objective is to use the number of packages of beef jerky sold ($X$, explanatory variable) to predict the number of jars of jam sold ($Y$, response variable).

  • The dataset of $n = 10$ paired sales transactions:

  • Jerky ($X$): $0, 8, 7, 8, 0, 9, 6, 8, 7, 6$

  • Jam ($Y$): $0, 5, 4, 7, 2, 4, 4, 6, 1, 7$

• Least Squares Regression Equation:

  • $\hat{Y} = 1.202 + 0.474X$

  • $\hat{Y}$ represents the predicted number of jars of jam sold.

  • $X$ represents the actual number of packages of beef jerky sold.

• Interpretation of the Intercept ($b_0 = 1.202$):

  • A customer who buys no beef jerky ($X = 0$) is predicted to buy $1.2$ jars of jam.

• Interpretation of the Slope ($b_1 = 0.474$):

  • For every additional package of beef jerky purchased, a customer is predicted to buy an additional $0.474$ jars of jam.

Residual Analysis in Linear Modeling

• Definition and Mathematical Formulation of Residuals:

  • A residual ($e$) measures the vertical prediction error between an observed data value and the corresponding predicted value generated by a regression equation:

  • $e = Y - \hat{Y}$

  • $Y$ denotes the actual observed response value.

  • $\hat{Y}$ denotes the predicted response value.

• Application to Customer Behavior Model:

  • The cheese shop models customer spending behavior using a regression model based on the amount of time spent inside the shop.

  • Observed transaction values:

  • Time spent in shop ($X$): $20\,\text{minutes}$

  • Actual customer expenditure ($Y$): $\$51.31$

  • Calculated Residual:

  • $e = -\$2.11$

  • Implied Predicted Spending ($\hat{Y}$):

  • $\hat{Y} = Y - e = 51.31 - (-2.11) = \$53.42$

• Contextual Interpretation of the Residual:

  • The customer spent $\$2.11$ less than one would predict based on the time spent in the shop.

Properties of Confidence Intervals: Point Estimates and Margin of Error

• Deconstructing a Confidence Interval:

  • Any symmetric two-sided confidence interval for a single parameter is centered around the point estimate and spans a distance equal to the margin of error in both directions:

  • $\text{Confidence Interval} = (\text{Point Estimate} - ME, \text{Point Estimate} + ME)$

• Weekly Cheese Sales Application:

  • A $99\%$ confidence interval for the total amount of cheese sold in a week at the shop is between $220.6\,\text{lb}$ and $250.8\,\text{lb}$.

  • Calculation of Point Estimate:

  • The point estimate is the arithmetic midpoint of the confidence interval:

  • $\text{Point Estimate} = \frac{\text{Upper Limit} + \text{Lower Limit}}{2} = \frac{250.8 + 220.6}{2} = \frac{471.4}{2} = 235.7\,\text{lb}$

  • Calculation of Margin of Error ($ME$):

  • The margin of error is the half-width of the interval:

  • $ME = \frac{\text{Upper Limit} - \text{Lower Limit}}{2} = \frac{250.8 - 220.6}{2} = \frac{30.2}{2} = 15.1\,\text{lb}$

• Impact of Confidence Level on Interval Width:

  • Scenario: Evaluating what happens if the confidence level is reduced from $99\%$ to $90\%$, holding sample size and sample variance constant.

  • Result: The interval would be narrower.

  • Mathematical Rationale:

  • Decreasing the confidence level decreases the required critical value ($z^*$ or $t^*$).

  • Because $ME = (\text{critical value}) \times SE$, a smaller critical value directly yields a smaller margin of error, which contracts the total width of the interval.