Day 2 lecture: Building dashboards in Tableau¶

This morning you learned what makes a chart and a dashboard clear. This afternoon you build them in Tableau: connecting to data, dimensions and measures, aggregation, the Marks card, filters, calculated fields, sheets, dashboards, actions, formatting and printing to PDF. Short Python cells show the same ideas in pandas, so you can see that both tools do the same arithmetic.

Open Tableau Desktop and this notebook side by side. Follow the numbered steps in Tableau; run the Python cells in JupyterLab with Shift + Enter.

What you will be able to do¶

1. Connect Tableau to a saved data source and to a CSV file, and check the data types.
2. Tell dimensions from measures and discrete (blue) from continuous (green) fields.
3. Explain how Tableau aggregates a measure for each mark, and change the aggregation.
4. Use the Marks card (Color, Size, Label, Detail, Tooltip) and the Filters shelf.
5. Write a calculated field, including a ratio KPI computed as a ratio of sums.
6. Assemble sheets into a dashboard, add a filter action, and format it for a reader.
7. Save a packaged workbook and print a dashboard to PDF for submission.

Reading: Tableau Software (n.d.), What is data visualization? and Tableau for students (licence and download).

Field names differ between versions

Tableau has changed its Sample – Superstore data several times: dates, field names (State or State/Province) and some values differ between versions. Use the names you see in your Data pane. When a CSV column is called total_bill, Tableau may display it as Total Bill; it is the same field.

import pandas as pd

pd.set_option("display.float_format", "{:,.2f}".format)
tips = pd.read_csv("data/tips.csv")
print("Rows (bills):", len(tips))
print("Columns:", list(tips.columns))

Rows (bills): 244
Columns: ['total_bill', 'tip', 'sex', 'smoker', 'day', 'time', 'size']

Part 1: The Tableau workspace¶

When you open a worksheet, the screen has the same parts every time:

The three icons next to the sheet tabs create a new worksheet, a new dashboard and a new story.

How Tableau thinks

You never write "draw a bar chart". You drag fields onto shelves, and Tableau decides the chart from what you placed where. A measure on Rows and a dimension on Columns gives a bar chart; a date on Columns gives a line. Show Me is a shortcut, not a different method.

Part 2: Connecting to data¶

The built-in sample¶

1. Open Tableau Desktop. On the start page, under Saved Data Sources (left side), click Sample – Superstore.
2. A blank worksheet, Sheet 1, opens. The Data pane lists tables such as Orders, People and Returns.

A CSV file¶

1. On the start page, under Connect → To a File, click Text file.
2. Go to the course folder notebooks/China_Classroom/data/ and open tips.csv.
3. The Data Source page shows a preview of the rows. Above each column an icon shows the data type Tableau guessed:

1. Click a type icon to change it if the guess is wrong. Then click the Sheet 1 tab at the bottom to start building.

Check what one row means

As in Python, look at the preview and say what one row is before you build anything. In tips.csv one row is one bill (one table served). In Superstore's Orders table one row is one order line (one product in one order), not one order.

The Python version of "connect and preview" is the cell you ran above, plus head():

tips.head()

Part 3: Dimensions and measures, discrete and continuous¶

Tableau sorts every field in two ways.

And separately, by colour of the pill:

Usually dimensions are discrete and measures continuous, but not always. A date can be either: YEAR(Order Date) as a blue pill gives one column per year; a green continuous month gives one long timeline. In Lab 2 you will change a date pill from discrete to continuous to draw a monthly trend.

A number that is really a category

In tips.csv, size (the number of people at the table) arrives as a measure, so Tableau would add it up. To compare party sizes, right-click size in the Data pane → Convert to Dimension, and Tableau treats each size as a group. In Python the same decision is whether you write groupby("size") or ["size"].sum().

Check your understanding

For each Superstore field, say dimension or measure, and discrete or continuous in a typical chart: Region, Sales, Order Date on a monthly trend line, Quantity, Customer Name.

Part 4: Aggregation and the level of detail¶

When you drag a measure into the view, Tableau does not draw one mark per row. It draws one mark for each combination of the dimensions in the view, and aggregates the measure inside each mark. The default aggregation is SUM.

In Tableau (with Sample – Superstore):

1. Drag Region to Columns and Sales to Rows. The pill reads SUM(Sales): one bar per region, each the total of its order lines.
2. Right-click the SUM(Sales) pill → Measure (Sum) → Average. The pill becomes AVG(Sales) and each bar shows the average order line, a different question.
3. Drag Segment to Color. Now there is one mark for each region and segment: the level of detail got finer, and each mark aggregates fewer rows.

The same three steps in pandas, on the tips data (grouping by smoker instead of Region):

# Step 1: SUM, one value per group (Tableau's default).
print("SUM(total_bill) by smoker:")
print(tips.groupby("smoker")["total_bill"].sum().round(2))

# Step 2: AVG, the same groups, a different question.
print("\nAVG(total_bill) by smoker:")
print(tips.groupby("smoker")["total_bill"].mean().round(2))

# Step 3: a finer level of detail, one value per smoker AND time.
print("\nSUM(total_bill) by smoker and time:")
print(tips.groupby(["smoker", "time"])["total_bill"].sum().round(2))

SUM(total_bill) by smoker:
smoker
No    2,897.43
Yes   1,930.34
Name: total_bill, dtype: float64

AVG(total_bill) by smoker:
smoker
No    19.19
Yes   20.76
Name: total_bill, dtype: float64

SUM(total_bill) by smoker and time:
smoker  time  
No      Dinner   2,130.14
        Lunch      767.29
Yes     Dinner   1,530.16
        Lunch      400.18
Name: total_bill, dtype: float64

Predict, then reveal

In Tableau you put Category on Rows and SUM(Sales) on Columns: 3 bars. Then you drag Order ID onto Detail. How many marks do you expect now, and why does the chart stop making sense? (Try it.)

Part 5: The Marks card¶

The Marks card controls how each mark looks.

Try it (Superstore): on a sheet with Sub-Category on Rows and SUM(Sales) on Columns, drag Profit onto Color. Tableau uses a diverging orange–blue palette centred on zero: orange for losses, blue for profits. This is this morning's diverging palette, chosen automatically because Profit can be negative. Click Color → Edit Colors to check where the centre sits.

Part 6: Filters¶

A filter limits which rows reach the view.

1. Drag a dimension (for example Region) to the Filters shelf, tick the values to keep, click OK.
2. Right-click the pill on the Filters shelf → Show Filter to put a filter control next to the chart for the reader.
3. To make one filter change several sheets: on the filter control's menu (small arrow) → Apply to Worksheets → All Using This Data Source (or Selected Worksheets).
4. Measures can be filtered too: drag Sales to Filters and choose a range of SUM(Sales).

In pandas, a filter is a boolean condition applied before grouping:

dinner = tips[tips["time"] == "Dinner"]          # filter: keep dinner bills only
print("Bills after the filter:", len(dinner), "of", len(tips))
dinner.groupby("smoker")["tip"].sum().round(2)  # then aggregate, as Tableau does

Bills after the filter: 176 of 244





smoker
No    331.45
Yes   214.62
Name: tip, dtype: float64

State the filter on the page

A filtered chart answers a narrower question. Put the filter in the title or the scope note ("Dinner bills only"), or the reader will assume the whole dataset.

Part 7: Calculated fields and a ratio KPI¶

A calculated field is a new field defined by a formula. Create one with Analysis → Create Calculated Field, or right-click in the Data pane → Create → Calculated Field. Give it a name, type the formula, and check that the message under the box says The calculation is valid.

Row-level versus aggregate calculations¶

This morning (Part 11) you saw that these answer different questions, and that a KPI tile should normally use the ratio of sums. In Tableau, the SUM(...) / SUM(...) form is how you write it. Superstore's standard KPI is:

Profit Ratio = SUM([Profit]) / SUM([Sales])

After creating it, right-click the field → Default Properties → Number Format → Percentage so it always displays as a percentage.

Here are both versions in pandas, for each value of smoker:

kpi = tips.groupby("smoker").agg(tips_sum=("tip", "sum"), bills_sum=("total_bill", "sum"))
kpi["ratio of sums %"] = 100 * kpi["tips_sum"] / kpi["bills_sum"]          # SUM([Tip]) / SUM([Total Bill])
kpi["mean of ratios %"] = 100 * (tips["tip"] / tips["total_bill"]).groupby(tips["smoker"]).mean()  # AVG([Tip] / [Total Bill])
kpi

The two percentage columns in the output above differ. Look closely: do they even rank smokers and non-smokers in the same order? If a dashboard tile said "Tip rate" without saying which one, two analysts could report different numbers from the same data. Name the formula in the tile or the scope note.

Other useful calculations

• IF SUM([Profit]) < 0 THEN "Loss" ELSE "Profit" END: a label you can put on Color.
• DATEDIFF('day', [Order Date], [Ship Date]): days to ship, per row.
• COUNTD([Order ID]): distinct orders, because Superstore has several rows per order.

Check your understanding

In Superstore, why does AVG([Profit] / [Sales]) give a different number from SUM([Profit]) / SUM([Sales]) for the whole company? Which one would you put on an executive dashboard, and why?

Part 8: From sheets to a dashboard¶

A dashboard is built from worksheets: build and title each chart on its own sheet first, then arrange them.

Plan before you build¶

Use this morning's principles. On paper, sketch:

1. The decision and the audience (for example: a regional manager deciding which products to review).
2. The 3–5 views, each answering one part of the question.
3. The layout: headline and KPIs at the top left, detail below and to the right.

Build¶

1. Build each sheet and rename its tab (double-click the tab), for example Sales trend, Profit by state, Sales by sub-category. The tab name becomes the default title on the dashboard.
2. Click the New Dashboard icon at the bottom.
3. On the left, under Size, choose Automatic so the dashboard fills the screen (or a Fixed size such as A4 landscape if you will print it).
4. Under Sheets, drag each sheet onto the canvas. Grey shaded areas show where it will land: top, bottom, left or right of what is already there.
5. Leave objects Tiled (the default) so they line up in a grid; Floating objects sit on top and are easy to misalign.
6. Add a Text object (under Objects) for the headline message and a small one for the scope note (period, data source, units).
7. Remove legends you do not need (click the legend → x), and delete repeated titles.

Check your understanding

Your dashboard has six sheets and three legends and still has no headline. Using this morning's principles, name three changes.

Part 9: Actions: making the dashboard interactive¶

Actions let a click on one view change the others.

The quick way: Use as Filter¶

1. On the dashboard, click once on a view (for example a map) to select it. A grey border appears with small icons at its top right.
2. Click the funnel icon (Use as Filter). Now clicking a mark in that view (a state) filters every other view on the dashboard.
3. To clear the selection, click the same mark again, or click an empty part of the view, or press Esc.

The full way: Dashboard → Actions¶

Dashboard → Actions → Add Action offers:

For each action choose the source sheet, target sheets, and whether it runs on Hover, Select (click) or Menu, and what happens when the selection is cleared (Show all values is usually what a reader expects).

In Python there is no clicking, but the idea of one filter driving every view is a function with a parameter. Every number below is recomputed for the chosen group:

def mini_dashboard(data, smoker=None):
    """Recompute every KPI for one selection, as a dashboard filter action does."""
    view = data if smoker is None else data[data["smoker"] == smoker]
    return pd.Series({
        "bills": len(view),
        "total spend (dollars)": view["total_bill"].sum(),
        "tip rate, sum/sum (%)": 100 * view["tip"].sum() / view["total_bill"].sum(),
        "average party size": view["size"].mean(),
    }).round(2)


pd.DataFrame({
    "All bills": mini_dashboard(tips),
    "Click: smoker = Yes": mini_dashboard(tips, "Yes"),
    "Click: smoker = No": mini_dashboard(tips, "No"),
})

Actions do not print

A PDF shows one state of the dashboard. Before printing, clear all selections so the reader sees the whole picture, unless the assignment asks for a filtered view; then say in a caption which state is selected.

Part 10: Formatting for the reader¶

Apply this morning's design rules inside Tableau:

Bars still start at zero in Tableau

Tableau's bar axes start at zero by default. If you right-click an axis → Edit Axis and untick Include zero, you have built this morning's misleading truncated bar chart.

Part 11: Saving, printing to PDF and publishing¶

Before you print, check

1. The dashboard tab is the active sheet.
2. No mark is selected and no filter is left on by accident.
3. Every title is a message, every axis has units, and the scope note is visible.
4. Open the PDF and read it as the reader will: is anything cut off or too small?

Summary¶

• Tableau draws one mark per combination of dimensions in the view and aggregates each measure in that mark (SUM by default), exactly like groupby in pandas.
• Blue pills are discrete (headers), green pills are continuous (axes). Dates can be either.
• The Marks card maps fields to colour, size, label, detail and tooltip; the Filters shelf limits rows.
• Ratio KPIs are calculated fields written as SUM([numerator]) / SUM([denominator]).
• A dashboard is a few titled sheets, laid out for one decision, connected by actions, with a headline and a scope note.
• Submit dashboards as a PDF from File → Print to PDF; keep a .twbx copy of your work.

Closing quiz¶

1. What is the difference between a dimension and a measure? Give one example of each from Superstore.
2. A pill is green. What does that tell you, and what will it draw?
3. You put SUM(Sales) on Rows and Region on Columns, then add Segment to Color. What happens to the number of marks?
4. Write the calculated field for a profit ratio KPI. Why not AVG([Profit] / [Sales])?
5. What does the funnel (Use as Filter) icon on a dashboard view do?
6. Name two things to check before Print to PDF.

Part 12: Assignment 1 briefing¶

Assignment 1: Data visualisation and descriptive analytics report (6%, CLO 1–3)

Due Sunday 15 November 2026, 23:59 China Standard Time, on GAP (Moodle).

You act as an analytics consultant for Bistro Analytics, advising a busy restaurant. Using the tips data (244 bills), you run a Python notebook that produces two summary tables and three figures, then write a short business report with four fixed sections:

1. Executive summary
2. Descriptive data overview
3. Visual insights and graphical trends
4. Strategic business recommendations

The restaurant owner wants help with two decisions: when to schedule the strongest staff and how to raise revenue and server income through upselling.

Everything you learned today applies: message titles, honest axes, ratio of sums, and turning findings into a recommendation for a specific audience.

Start here: Assignment 1 brief and starter notebook. Open D2_assignment1_tips_report.ipynb in JupyterLab, read the brief, and save your own copy as YourLastName_DATA300_Assignment1.ipynb.

After this briefing, the lab starts with Applied Lab 2: Dashboard creation. If you finish the lab early, start Assignment 1.