Lecture 3: Master the Key Figures

STA1000 - Data Analytics and Metrics

Damien Dupré

Previously …

In Lecture 2 we opened insta-data.csv, cleaned and transformed data, and identified the type of each variable: categorical or continuous.
We also met Tableau: Data Source, Worksheet and Dashboard, with Dimensions for categorical variables and Measures for continuous ones.

Today it is time to look at the data: which figure for which variables, and how to build it in Tableau.

Why Use Figures?

Data visualisations are important:

For the analyst, to understand what is going on in the data (errors, unexpected values, analyses to run)
For the reader, to access the message and be convinced: a report tells a story, and each figure supports it

Figures can also lie or mislead: use them with ethics and professionalism.

1. The Key Figures

What Would You Like to Show?

Based on Andrew Abela’s Chart Chooser: start from the question, not from the chart

Keep Your Figures Clear and Self-Explanatory

1. Clear

Complex figures should be crystal clear: use design features (colours, sizes, …) to make them clear

2. Self-Explanatory

Always use a title, a legend and well-defined axes

Source: www.venngage.com

Figures Aggregate Data

A figure aggregates data into information with basic statistical calculations

Student	Group	Mark
S1	A	60
S2	B	70
S3	A	60
S4	B	55
S5	A	65
S6	B	65
S7	A	65

Count: A = 4, B = 3
Proportion: A = 57%, B = 43%
Density: 55 = 14%, 60 = 29%, 65 = 43%, 70 = 14%
Median and quartiles: Q1 = 60, Median = 65, Q3 = 65
Mean and Standard Deviation: Mean = 62.9, SD = 4.9

Zoom on Median and Quartiles

Sort the values, then cut them into four groups of equal size. The interquartile range (IQR = Q3 - Q1) holds the middle 50% of the data.

Zoom on the Box Plot

The box runs from Q1 to Q3, and the line inside is the median
The whiskers stop at the most extreme values within 1.5 × IQR of the box
Anything further away is drawn as a dot: a possible outlier

Zoom on Mean and Standard Deviation

Mean (or average) is the sum of all the values divided by their number
Standard deviation (SD) is the typical distance to the mean: the square root of the average squared distance

Source: Cuemath

Zoom on Regression

The line that best fits the points of two continuous variables is a regression line: Y = b₀ + b₁X + e

b₀ is the intercept: the value of Y when X is 0
b₁ is the slope: how much Y changes when X increases by 1
e is the error (or residual): the distance between each point and the line

Zoom on Correlation

The correlation (r) is the slope of the regression line once both variables are on the same scale. It measures how closely the points follow a straight line, from -1 to +1:

Which Figure for Which Variables?

In Tableau, categorical = Dimension and continuous = Measure.

Your Turn!

Which family of figures?

Variable(s) in `insta-data.csv`	Family
`format`
`reach`
`format` and `reach`
`format` and `is_paid`
`impressions` and `likes`
`theme`

Which Family of Figures?

Variable(s) in `insta-data.csv`	Family	Figure
`format`	Composition	Bar chart
`reach`	Distribution	Histogram, box plot
`format` and `reach`	Comparison	Box plots
`format` and `is_paid`	Comparison	Stacked bars
`impressions` and `likes`	Relationship	Scatterplot
`theme`	Composition	Bar chart, not a pie

2. The Four Families on Loomrow

Composition Figures

Count of the different categories

Composition Figures

Proportion of the different categories

When to Use a Pie Chart?

Only when the variable has two categories

With more categories, use a bar chart

Source: Depict Data Studio

When to Use a Pie Chart?

Here is a pie chart that works:

Distribution Figures

Four figures of one continuous variable, reach

Each figure keeps something and throws the rest away: the dynamite keeps only the mean.

Mean or Median?

Reach is right-skewed: most posts reach a few thousand accounts, a few reach tens of thousands, and one reel reached 535,086.

The mean (9,788) is pulled towards that tail. The median (5,184) is the typical post.

Mean ± SD = 9,788 ± 27,748: a range that goes below zero, for a count that cannot be negative.

Comparison Figures

Two categorical variables: format and is_paid

Comparison Figures

A categorical and a continuous variable: format and reach

Relationship Figures

Two continuous variables: impressions and likes

Relationship Figures

Adding a categorical variable: one panel per format, same axes

For the same impressions, carousels earn about 6 likes per 100 impressions and reels fewer than 3.

3. The Key Figures in Tableau

Variable Types and Tableau Figures

A family of figures for each type of variable:

Composition: 1 Dimension
Distribution: 1 Measure
Comparison: 2 Dimensions, or 1 Dimension and 1 Measure
Relationship: 2 Measures

Show Me highlights the figures they allow, it also explicitely indicate what combinaison of Dimension and Measure is needed.

Composition in Tableau

A Dimension on Columns and the count field on Rows. For percentages, open the menu of the count pill (click its arrow, or right-click / ⌃ Control + click) → Quick Table Calculation → Percent of Total.

The recordings in this section use another dataset: the clicks are the same with insta-data.csv.

Composition in Tableau: Pie Chart

Only for variables with two categories: the Dimension on Columns, the count on Rows, then Show Me → pie chart.

Distribution in Tableau

Tableau does not draw every distribution figure:

Histogram: yes, with Show Me
Density: no density curve out of the box
Box plot: yes, with Show Me or the Analytics pane
Dynamite: an average bar, but no error bars

Distribution in Tableau: Histogram

A Measure on Rows, then Show Me → histogram.
Tableau creates a new field, variable (bin): to change the width of the bars, open its menu → Edit.

Distribution in Tableau: Box Plot

A Measure on Rows, then Show Me → box-plot
Use dimension to display all the values and the 2 whiskers.

Distribution in Tableau: Average Bar

A Measure on Rows is a SUM by default: open its menu → Measure → Average.
Alternatively, the same menu offers Median if needed.

Your Turn!

Composition and distribution in Tableau

With insta-data.csv, one worksheet each:

Posts per format: count, then percent of total
A pie chart of is_paid
A histogram of reach: where are most posts?
A box plot of reach, post_id on Detail: which post is the highest dot?
reach as an Average, then as a Median: which would you report?

Comparison in Tableau

Add a Dimension on Colour, Columns or Rows to any of the previous figures: bar or pie charts with the count, histograms, box plots, bar charts with the average.

Comparison in Tableau: Multiple Bars

A second Dimension on Columns puts the bars side by side; on Colour, it tells the groups apart.

Comparing Distributions in Tableau

A Dimension on Rows: one histogram per group

A Dimension on Columns: one box per group

Relationship in Tableau

The scatterplot is the best way to show the relationship between 2 Measures, but:

switch both from the default SUM to Dimension (pill menu), or Tableau draws a single dot for the sum of everything
add a Trend Line from the Analytics pane

Relationship in Tableau: Scatterplot

Two Measures on Columns and Rows, each switched to Dimension. Analytics pane → Trend Line → Linear. A Dimension on Colour gives one line per group: hover a line to read its R² and p-value.

Your Turn!

Comparison and relationship in Tableau

Number of posts per format, with is_paid on Colour
Box plots of reach by format: which format has the highest median?
Average reach by format: which format wins now, and why?
A scatterplot of impressions against likes with a trend line, then format on Colour: which format earns the most likes per impression?

Save the workbook to your Tableau Public profile: Lecture 5 turns these sheets into a dashboard.

Thanks for your attention and don’t hesitate to ask if you have any questions!

@damien_dupre

@damien-dupre

https://damien-dupre.github.io

damien.dupre@dcu.ie