Analysis of Unusual Values and Patterns in Scatter Plots

Overview of Scatter Plot Analysis

• Scatter plots display bivariate data to visualize the relationship, association, and dependency between two quantitative variables, typically placing an explanatory variable on the $x$-axis and a response variable on the $y$-axis.

• A rigorous evaluation requires examining four key characteristics of bivariate data:

  • Form: The overall shape of the relationship (e.g., linear, curved/non-linear, or clustered).

  • Direction: The overall trend indicating a positive association (as $x$ increases, $y$ tends to increase) or negative association (as $x$ increases, $y$ tends to decrease).

  • Strength: The degree of dispersion of data points around the underlying trend line or model curve (often measured by correlation coefficients such as Pearson's $r$).

  • Unusual Features: Data points falling outside the overall pattern, including clusters, gaps, and outliers.

• Complete exploratory analysis mandates looking beyond global summaries to distinguish between systematic structural behaviors and isolated individual anomalies.

Systematic Patterns and Weight Discrepancies

• Systematic Bias vs. Random Error:

  • Systematic Pattern: Occurs when residuals ($\text{residual} = y - \hat{y}$, representing the difference between observed value $y$ and predicted value $\hat{y}$) display a non-random distribution or consistent sign. Predictions that consistently exceed actual measurements reflect negative residuals ($y < \hat{y}$).

  • Random Error: Unpredictable, unsystematic scatter centered around a mean of zero ($\mu_{\epsilon} = 0$), naturally occurring due to inherent biological or measurement variability.

• Application to Weight Modeling (Otter Offspring Case):

  • In data modeling involving otters, calculated model values were found to be larger than an otter's actual weight.

  • A systematic pattern emerged where the model consistently overestimated the weight of daughters.

  • Consistent overestimation indicates a persistent, directional bias rather than chance variation, pointing to factors such as sexual dimorphism (e.g., females having lower average weights than a pooled population mean) or an uncalibrated regression slope.

Identification of Unusual Values

• Distinguishing Anomalies from Trends:

  • Scatter plot analysis requires explicitly separating pervasive model bias from isolated, individual data anomalies.

  • An outlier in bivariate data is an individual observation that falls notably far from the overall trend or regression line, exhibiting an unusually large vertical residual.

  • A high-leverage point possesses an extreme explanatory variable value ($x$-value) relative to the mean explanatory value, having the potential to exert substantial leverage on the regression line.

  • An influential point markedly alters the slope, intercept, or correlation coefficient when included versus omitted from the data analysis.

• Detection Techniques:

  • Plotting residuals against fitted values ($\hat{y}$) or the explanatory variable ($x$) to detect points that fall outside horizontal tolerance bands.

  • Identifying standardized residuals with an absolute magnitude exceeding threshold limits (such as $|z| > 2$ or $|z| > 3$).

Questions & Discussion

• Audience Inquiry: An opportunity for clarification was provided specifically concerning scenarios where the calculated model value exceeded an otter's actual weight ("Questions?").

• Resolution & Progression: No questions or objections were raised by the audience, permitting the presentation to progress directly from systematic trends into the identification and classification of unusual values ("Okay").