Superset vs Power BI, and data storytelling¶

Exercise comparison-and-storytelling-annex.pdf · PDF

Three friends have spent two years making a game. They are about to launch it on Steam, and they ask you, the one who "knows about data", a simple question: should we, and at what price?

You open your Steam market dashboard and turn the screen towards them. Four KPIs, a line, some bars, two filters. They look at it for ten seconds. Then one of them says: "Okay... so what do we do?"

Your dashboard is correct. Every number on it is true. And it did not answer the question. This last lesson is about that gap: first you compare the two tools you now know, and then you learn the skill that neither tool has a button for — saying what the data mean for the person in front of you.

Two tools, side by side¶

You have built the same three charts on two consoles. Time to compare them properly. A fair comparison does not start with "which is better". It starts with the same questions asked of both.

Two words in that table deserve a definition:

Open source software publishes its source code under a licence that lets anyone use it, study it, change it and share it. Proprietary software belongs to a company that keeps the code and sets the conditions of use.

So which one is better?¶

Refuse the question. Better for whom?

Same question, opposite answers, both right. The tool follows the context. Keep that word in your hand: you need it again in a minute.

From minimap to quest marker¶

All year, this module has been a crafting video game: you farmed the materials, smelted them, crafted with them, and built the HUD. A dashboard is the minimap of that HUD. It shows everything: every road, every enemy, every chest.

Now think of a new player dropped into that world. The minimap is complete, accurate and useless to them, because it does not say where to go. What they need is the quest marker: the little arrow that says there.

Data storytelling is turning a finding in the data into a message for a specific reader: what you found, why it matters to them, and what they should do about it.

A dashboard is the minimap: it shows everything. A story is the quest marker: it says where to go.

Your three friends looked at a perfect minimap and asked for the marker.

The three parts of a data story¶

A data story has three parts. Each one answers one question:

flowchart LR
    C["<b>Context</b><br>who reads, what they<br>must decide"] --> I["<b>Insight</b><br>one finding, with its<br>number and its 'so what'"]
    I --> A["<b>Action</b><br>what to do: starts<br>with a verb"]
    L["<b>Limit</b><br>what the data<br>cannot tell you"] -. "written next to" .-> I

The order matters, and it starts with the reader, not with the chart. This is the first lesson of one of the best-known books on the subject, Storytelling with Data by Cole Nussbaumer Knaflic (Wiley, 2015): before you draw anything, know who your audience is and what you need them to know or do.

That is also why a story for the studio of three is not the story for Steam's own managers. Same data, same dashboard, another quest.

What makes an insight¶

An insight is a finding that carries a number and a consequence: it changes what the reader decides.

Most first attempts at an insight are really a description: they say what the chart is about. The test is simple. Is there a number? Is there a "so what"?

Every number in the right-hand column comes from games_clean. Here is the first one, drawn:

%%{init: {"themeVariables": {"xyChart": {"plotColorPalette": "#4051b5"}}}}%%
xychart-beta
    title "Games released on Steam per year"
    x-axis ["2013", "2014", "2015", "2016", "2017", "2018"]
    y-axis "Games" 0 --> 9000
    bar [556, 1352, 2544, 4503, 6871, 8845]

From 1,352 games in 2014 to 8,845 in 2018. Call the chart "Releases per year" and the reader has to do that sum alone. Call it "Six times more games in four years" and the title has already done it.

The honest part: say the limit¶

An insight is a claim. A professional writes, next to it, what the data cannot tell.

A limit is one thing your data cannot show, stated next to the claim it affects.

Two limits sit inside the Steam data, and both would hurt your friends if you hid them.

The 2018 trap. The table says that 87.6% of the games released in 2018 are niche: fewer than 20,000 owners. Terrifying. But the data were collected in 2019. A game released in December 2018 had only weeks to find its players; a game from 2014 had years. 2018 holds a third of all the games and only 37 of the 623 hits. Its games are not worse. They are younger. The 87.6% is true and overstated at the same time.

Correlation is not cause. Look at the share of hits per price band:

"Cheap games fail, so charge 30." Careful. The table shows who is a hit, not why. Big studios charge more and have marketing budgets, famous names and hundreds of people. The price did not make the hit. Write it as a limit, not as advice.

Try it: the Steam story¶

Your reader: a studio of three people deciding whether to launch on Steam, and at what price. You will turn your Steam market dashboard from a minimap into a quest marker.

Step 1 — Find the insight. In SQL Lab, database Dades classe, run these two queries one at a time and read the results slowly:

SELECT release_year, COUNT(*) AS games,
       ROUND(100.0 * COUNT(*) FILTER (WHERE owners_tier = 'Niche') / COUNT(*), 1) AS niche_pct,
       PERCENTILE_CONT(0.5) WITHIN GROUP (ORDER BY price) AS median_price
FROM games_clean WHERE release_year >= 2013
GROUP BY release_year ORDER BY release_year;

SELECT price_band, COUNT(*) AS games,
       ROUND(100.0 * COUNT(*) FILTER (WHERE owners_tier = 'Hit') / COUNT(*), 1) AS hit_pct
FROM games_clean GROUP BY price_band ORDER BY hit_pct DESC;

The second query returns the price-band table you saw in the previous section. Now write one sentence with at least two numbers that would worry or encourage the studio.

Step 2 — Write the story. Three sentences, one per part. The insight must carry its numbers. The action must start with a verb. Give them to a classmate: if they can tell which sentence is which without being told, you are done.

Step 3 — Annotate the dashboard. A story nobody can see on the dashboard is a story only you know.

You are finished when someone who reads only the header and the two text blocks knows what the studio should do.

Key takeaways¶

• Superset is open source (Apache License 2.0), runs on a server you look after and reads one dataset per chart. Power BI is proprietary (Microsoft), builds on Windows, shares through Microsoft's cloud and imports the data into a semantic model.
• Neither is "better". The tool follows the context: who will use it, what they already have, who looks after it. A free licence is not a free tool.
• A dashboard is the minimap; data storytelling adds the quest marker. The data do not speak for themselves.
• A data story has three parts: context (who reads, what they decide), insight (one finding with its number and its "so what") and action (what to do, starting with a verb).
• A description is not an insight. "Releases per year" is a label; "releases multiplied by 6.5 in four years" is a finding.
• Always state a limit next to the claim it affects: the 2018 games are younger, not worse; a higher price goes with hits, it does not cause them.

Quick check¶

That was the last quest. You started this course looking at four datasets that hid behind identical statistics, and you finish it able to extract data, clean it, draw it and say what it means, limits included. Go back to the quest log, Learning outcomes, and read the nine criteria again: this time, every one of them describes something you have already done.