Graph Selection and Types

Graph Selection

• Consider perception, comparison, interpretation, and summarization when choosing a graph
• Perception involves how the graph is perceived
• Comparison involves making comparisons using the graph
• Interpretation considers what "larger is better" means in the graph
• Comprehension involves summarizing the information

Bar Chart

• Compares a set of values across different categories
• Easy to understand, good for comparison, and flexible
• Can be crowded if there are too many bars
• Not suitable for showing trends
• Needs careful ordering
• Used for comparing values across categories
• Should not be used for showing parts of a whole
• Example: Comparing sales across different product types

Clustered Bar Chart

• Compares values from multiple groups
• Provides clear group comparison and a familiar layout
• Effective for a small number of groups
• Can be confusing with too many groups
• Requires a legend
• Takes up a lot of space
• Example: Showing male vs. female performance across departments

Connected Dot Plot (Dumbbell/DNA Chart)

• Shows the change between two points
• Highlights change neatly, is compact, and less cluttered
• Can be tricky for many points and requires labels
• Not intuitive
• Used for showing before/after comparisons
• Should not be used for very large datasets
• Example: Change in customer satisfaction scores before and after a service redesign

Proportional Shape Chart

• Uses sized shapes to represent parts of a whole
• Eye-catching, good for visual storytelling, and shows proportions
• Hard to compare accurately and needs annotation
• Area can mislead
• Used for showing parts of a whole where precision is not critical
• Should not be used when precise comparison is needed

Bubble Chart

• Plots three or more variables (X, Y, size)
• Provides visual richness, shows clusters, and is dynamic
• Can be hard to read sizes, crowded, and color/size can mislead
• Used for showing relationships among multiple variables
• Should not be used for two-variable simple comparisons
• Example: Showing GDP, population, and CO2 emissions across countries

Radar Chart (Spidergram)

• Compares multiple variables radiating from the center
• Great for skill/attribute comparison and highlights outliers
• Overlapping lines can be hard to read and the scale can be confusing
• Used for profile comparison
• Should not be used for linear data
• Example: Comparing player performance across multiple skills

Polar Chart

• Radial graph for cyclic data
• Good for circular patterns and engaging
• Poor comparison across distance and nonlinear scaling
• Used for showing seasonal/cyclic data
• Should not be used for non-cyclic relationships
• Example: Monthly sales trends shown in a circular format

Range Chart

• Shows the minimum and maximum values for categories
• Highlights variability, simple, and easy
• Does not show distribution and can be confusing with many items
• Used for showing value ranges
• Should not be used for showing whole distribution
• Example: Showing temperature ranges across cities

Box and Whisker Plot

• Displays distribution with median, quartiles, and outliers
• Spots outliers quickly, summarizes distributions, and is compact
• Hard for beginners, no exact numbers, and less intuitive
• Used for showing variation across groups
• Should not be used for audiences unfamiliar with boxplots
• Example: Income distribution across regions

Histogram

• Binned frequency distribution
• Shows distribution shape, highlights skewness, and is easy to spot peaks
• Sensitive to bin width and poor category comparison
• Used for showing spread of continuous data
• Should not be used for category comparisons
• Example: Distribution of students' test scores

Word Cloud

• Size of words indicates frequency
• Good for qualitative data, has visual impact, and is engaging
• Poor for detailed analysis and hard to compare exactly
• Used for highlighting the most common words
• Should not be used for detailed text analytics
• Example: Most common keywords in customer feedback

Pie Chart

• Circle divided into slices representing parts
• Familiar, good with few categories, and easy to grasp
• Bad for close comparisons and easily misinterpreted
• Used for parts of a whole with 2-5 slices
• Should not be used for many small categories
• Example: Share of different expense categories in a budget

Stacked Bar Chart

• Bars divided into sections
• Shows composition, totals visible, and is compact
• Hard to compare internal parts and relies on color
• Used for showing overall totals and compositions
• Should not be used for too many small components
• Example: Market share by company and region

Back-to-Back Bar Chart

• Two bar charts facing opposite directions
• Shows demographic splits, efficient, and neat
• Can be tricky if not balanced and labels are critical
• Used for comparing two opposing groups
• Should not be used for multi-group comparisons
• Example: Male vs. female age distributions

Treemap

• Rectangles sized by value
• Space-efficient, highlights large items, and great for big datasets
• Tiny blocks unreadable and positioning is meaningless
• Used for visualizing parts of a large whole
• Should not be used for precise comparisons
• Example: Market share by company in a tech sector

Sunburst Chart

• Circular treemap showing hierarchy
• Clear hierarchy, attractive design, and engages viewers
• Tiny slices at edges and tricky comparisons
• Used for visualizing category breakdowns
• Should not be used for non-hierarchical data
• Example: Website navigation paths

Bubble Chart (Rich Version)

• X, Y, size, and color plotted simultaneously
• Multi-dimensional, cluster finding, and has visual impact
• Size is hard to judge and overlapping can occur
• Used for displaying complex multi-variable data
• Should not be used for simple comparisons
• Example: Global inequality vs. GDP vs. population

Chord Chart

• Circular flow between categories
• Many-to-many relations and elegant
• Hard for beginners and crowded easily
• Used for showing flows between categories
• Should not be used for simple one-way relationships
• Example: Migration flows between countries

Sankey Diagram

• Flow diagrams showing magnitude
• Great for proportional flows and storytelling power
• Hard to design and risk of crowding
• Used for showing energy use and money flows
• Should not be used for basic single-step processes
• Example: Energy input/output in a power plant

Line Chart

• Connects points to show trend over time
• Clear for trends, simple, and clean
• Risk of overlap
• Used for showing trend over time
• Should not be used for unordered categories
• Example: Revenue growth over 10 years

Gantt Chart

• A project schedule visualization
• Tracks project timelines and clear duration view
• Messy for large projects and constant updating required
• Used for managing projects
• Should not be used for small unrelated tasks
• Example: Construction project timelines

Piled Bars

• Wrapping bars into multiple columns
• Handles large datasets
• Organized visually
• Hard cross-comparison and breaks flow
• Used for displaying hundreds of categories
• Avoid use for small data
• Example: Survey results with 300+ options

Zvinca Plot

• Advanced scatter for distributions
• Large dataset summarization and unique
• Hard to interpret and nonstandard
• Used for very large distribution datasets
• Avoid use for simple distributions
• Example: Distribution of 1 million customer ratings

Linear Regression

• Used to estimate and model the relationships between variables
• Dependent Variable (Y) is the outcome to predict
• Independent Variables (X1, X2, ..., Xp) are the inputs
• Understand the relationship between variables (descriptive analytics)
• Predict future values (predictive analytics)

Simple Linear Regression

• Models the relationship between an independent variable (X) and a dependent variable (Y)
• Goal is to find a straight line that best fits the data points
• Equation of the line: Y = mX + c
• Example: Data points (X: 1, 2, 5, 3, 4; Y: 2, 4, 10, 6, 8), Formula learned: Y=2X

Least Squares Method

• Finds the best-fitting line by minimizing the sum of squared errors
• Tool in Excel: Data → Data Analysis → Regression

Multiple Linear Regression

• Models the relationship between several independent variables and one dependent variable
• Equation: Y = b0 + b1X1 + b2X2 + ... + bpXp
• Example output: Q = 205 - 3.9P + 0.6I + 1.33T, where:
  • Q = Quantity (dependent variable)
  • P = Price, I = Income, T = Temperature (independent variables)

Regression Model Check

• Look at the p-value: if p-value < 0.05, the model is statistically significant
• Check individual variables: if a predictor's p-value > 0.05, it may not be useful

Linear vs Logistic Regression

• Linear Regression:
  • Output is continuous (e.g., price, temperature)
  • Predicts a real number
  • Example: Predict a salary
• Logistic Regression:
  • Output is categorical (e.g., pass/fail, yes/no)
  • Predicts a probability (0 to 1)
  • Example: Predict if a customer buys or not

Linear Regression Use Cases

• Simple, easy to understand and apply
• Good for data with a linear relationship and strong relationships
• Provides a clear mathematical formula
• Forecasting sales based on advertising spend
• Predict house prices based on size, location, and number of bedrooms
• Estimating students' final grades
• Modeling relationship between height and weight

Linear Regression Bad Use Cases

• Predicting with categorical variables only
• Predicting pass/fail outcomes
• Modeling non-linear relationships
• When data has many outliers or is highly skewed

Linear Regression in Excel

• Go to Data → Data Analysis → Regression
• Set Input Y Range (dependent variable) and Input X Range (independent variable[s])
• Check p-values for significance and coefficients to build your formula

Regression Statistics Metrics

Multiple R

• Correlation coefficient between predicted and actual values
• Values between -1 and 1 indicate the strength of the relationship

R Square

• Coefficient of determination
• Shows the variance in Y that is explained by the X variables.

Adjusted R Square

• Adjusts R² for the number of predictors
• Prevents overestimating model quality

Standard Error

• Average distance that the observed values fall from the regression line
• Lower is better

Observations

• The number of data points used in the analysis

ANOVA Table Components

• df (degrees of freedom): Number of values that can vary, Regression df = number of predictors
• SS (Sum of Squares): Measures total variation, Divided into Regression SS (explained) and Residual SS (unexplained)
• MS (Mean Square): SS divided by df, Used to calculate the F-statistic
• F: F-statistic for model fit, Higher F = more likely at least one predictor is useful
• Significance F: p-value for the F-test, If Significance F < 0.05, the model is statistically significant

Coefficients Table

• Coefficients: Numbers you plug into the regression equation
• Standard Error: Accuracy of each coefficient estimate, Smaller = more precise estimate
• t Stat: Tests whether the coefficient is significantly different from zero, Higher absolute value = more significant
• P-value: Probability that the coefficient is actually zero (no effect), If p-value < 0.05, that variable is statistically significant
• Lower 95% and Upper 95%: 95% confidence interval for each coefficient,True coefficient is expected to fall within this range 95% of the time

Data Analytics Pipeline

• Acquire Data: Collect data from various sources like CSV files, web scraping, internal/external systems, open data platforms, always check data formats and licensing.
• Clean Data, Normalise Data, Impute if Necessary
  • Clean removes duplicates, fixes incorrect entries, standardizes formats
  • Normalise ensures one value per cell, creates primary keys, and enforces data types
  • Imputation fills missing values

Statistical Understanding

• Familiarise with Data by running descriptive statistics and looking at distributions
• Develop Instinct or Intuition by answering: "What is unusual here?". "What factors might influence this outcome?".
• Formulate a hypothesis: Higher seat rows correlate with lower Year 1 scores
• Apply reasoning through: Correlations, regressions, t-tests, chi-square tests to see of these assumptions hold true
• Visualise data using bar charts (categories), line charts (trends), Scatter plots (relationships), Histograms (distribution shapes), and always pick the simplest chart
• Write the Narrative in a clear story
• Conclude to prioritise most important findings to guide future decisions