Built-in AI tutor. Stuck on how to test a claim against the data? Ask the helper on this page. It coaches, it does not hand you the answer.
A chart is not a decoration. It is an argument, and arguments can
be wrong, slanted, or honestly mistaken. This page is about reading a chart the way
an analyst should: as a claim someone is making, that you are responsible for testing.
A chart is an argument: claim, evidence, read
Every good chart makes one point. Pull it apart into three pieces and you can
judge it instead of just nodding at it.
Claim
The one sentence the chart is meant to support. "Commercial customers tie up most of our overdue cash."
Evidence
The actual data drawn: which rows, which measure, which time window, which segment.
Read
What the picture honestly shows. Not what you hoped it would show, and not more than the data can carry.
When the read matches the claim and the evidence is complete, the argument holds.
When the claim runs ahead of the evidence, you have found something worth questioning.
Edward Tufte calls the standard here graphical integrity: the picture
a chart shows should match the numbers underneath it, with nothing exaggerated and
nothing hidden.
Correlation is not causation
Two things moving together does not mean one caused the other. Late-paying
customers might also be your largest customers, so a chart linking "size" and
"lateness" could be picking up a third thing entirely, like the net terms you
gave them. A trendline that slopes up tells you the two measures move together.
It does not tell you why, and it does not give you permission to say one drives
the other.
Before you write "X causes Y," ask what else could produce the
same picture. If you can name a plausible third factor, the chart has not earned
the causal claim.
The biases that derail the read
The hardest errors are not in the data. They are in us. Daniel Kahneman, in
Thinking, Fast and Slow (2011), describes how quick, intuitive judgment
runs ahead of slow, deliberate checking, and how that gap is where bias lives.
Three patterns show up constantly in chart-reading.
Confirmation bias. We reach for the chart that confirms
what we already believe, and we skim past the one that does not. If you walked in
sure that Consumer customers are the problem, you will notice the Consumer pie and
forget to check the dollars.
Motivated reasoning. When a conclusion is one we want to
be true, we hold it to a lower standard and we rationalize the gaps. The well
documented finding is that people reason toward conclusions they are already
motivated to reach, then build the justification afterward. The chart becomes a
prop for a decision already made.
Cherry-picking. Pick the time window where the line goes
your way. Pick the segment that supports the point. Drop the bucket that does not.
Each choice can be defended on its own, but together they manufacture a conclusion
the full data would not support.
The rationalization trap: start with the answer, and the data will agree
The most dangerous failure in business analytics is not a bad chart. It is a
competent analyst who already knows what the answer is supposed to be. Once you
start from an expectation, the analysis quietly stops being a test and becomes a
confirmation. The mechanism has three steps, and none of them feels dishonest
while it is happening.
Expect
You arrive with the conclusion.
Your manager said it, last quarter showed it, or you just believe it. You have not
decided to be biased. You simply expect a certain answer.
Filter
Your scrutiny turns asymmetric.
Of evidence that fits the expectation you ask "Can I believe this?",
and almost anything passes. Of evidence that contradicts it you ask
"Must I believe this?", and you hunt for a reason to set it
aside. Same data, two different standards, applied without noticing.
Justify
You end exactly where you
started, except now you are holding a chart. The expectation has become a
conclusion with an exhibit attached, and it feels like analysis because real work
happened in between.
This is why it is so dangerous for a business. Rationalized analysis does not
look like a lie; it looks like diligence. The collections team chases the segment
everyone already blames while the real exposure ages quietly. A forecast is built
to justify a deal someone already wants to sign. A write-down is deferred because
every review started from "we would rather this were fine." The chart gives a
pre-made decision an audit trail it never earned, and the organization finds out
the truth on the schedule of the data, not on its own.
The defense is not to have no expectations. You will always have
them. The defense is to write your read of the data before
you reread the claim, and to notice which question, "can I" or "must I," you are
asking of each chart.
Feel it happen: read the chart twice
This exercise works only if you play it straight. Read the memo, look at the
chart it points you to, and write your one-sentence read before going on. The
data here is a synthetic write-off series, not the homework data.
From your manager, Monday 8:04 AM: "Consumer write-offs
are out of control again this quarter. Pull the chart and confirm so I can take
it to the controller."
Write-offs by segment, $ thousands. The quarter-over-quarter view
attached to the memo.
Now read it again without the memo. Here is the
same series with the window the memo left out.
Write-offs by segment, $ thousands, all eight quarters.
The full window tells a different story. Consumer spikes every fourth quarter,
35 last year and 36 this year, then returns to its low-20s level: that is
seasonality, not a segment out of control. Meanwhile Commercial has climbed from
38 to 72 and now carries roughly twice Consumer's dollars. If the company spends
its collections energy on Consumer because the memo said so, the exposure that is
actually growing keeps growing.
Look back at the sentence you locked in. Did it mention Commercial at all? Most
readers' first sentence echoes the memo: the expectation chose what you looked at,
and the two-quarter window made the expectation easy to confirm. Nothing about the
zoomed chart was false. It was just the view that made "can I believe the memo?"
the only question you asked.
With the full window visible, which is the honest read?
The Q4 spike is real but repeats annually at the same size, while Commercial's climb is steady and large. "Nothing to act on" overcorrects: a doubling trend in the bigger segment is exactly what to escalate.
You wrote your first read with the memo in front of you. The trap to take away is:
The zoomed chart was accurate and managers will always arrive with expectations. The failure mode is asymmetric scrutiny. The fix is procedural: write your own read of the full data before rereading the claim.
This is exactly the homework. Two leaders hand you their
expectations about Summit Gear's receivables. Write your own read of the dashboard
first, then hold each claim against it and ask "must I believe this?" of both.
What a chart shows vs. what a leader wants it to show
A senior leader will sometimes arrive with the conclusion first and ask you to
find the chart for it. That is normal, and it is not necessarily dishonest. Often
they genuinely believe it. Your job is not to argue with the person. Your job is
to check whether the data supports the claim, and to say so plainly when it does
not.
Watch for the gap. The claim is about all overdue cash; the chart shows only
one quarter. The claim is about every customer; the chart filtered to one segment.
The claim says "most"; the chart shows 40 percent. Naming the gap is the whole
skill. You are not accusing anyone. You are reporting what the evidence will and
will not carry.
In the homework you will do exactly this. A leader makes a
claim about Summit Gear's receivables. You hold it against the data and decide
whether the chart actually shows what the claim needs it to show.
A checklist for stress-testing a claim against data
Run a claim through these before you repeat it or act on it.
State the claim in one sentence. If you cannot, the argument is not clear enough to test.
Check the measure. Is this counting invoices or summing dollars? The two often disagree.
Check the window. Why this date range? Would a longer or shorter one flip the read?
Check the population. Which segment, which customers, which rows were included or dropped?
Check for a third factor. Could something else explain the pattern just as well?
Look for the disconfirming chart. What view would prove the claim wrong, and have you drawn it?
Match read to claim. Does the picture support the exact sentence, or only a weaker version of it?
If a claim survives all seven, you can stand behind it. If it fails
one, you have found the question to ask out loud.
Sources
Daniel Kahneman, Thinking, Fast and Slow (Farrar, Straus and Giroux,
2011). On fast intuition outrunning slow checking, and the roots of confirmation
bias and motivated reasoning.
Edward R. Tufte, The Visual Display of Quantitative Information
(Graphics Press). On graphical integrity: a chart should show the numbers truthfully.
The documented finding that people tend to reason toward conclusions they are
already motivated to reach, then construct justifications afterward. See work on
motivated reasoning summarized by Ziva Kunda, "The Case for Motivated Reasoning,"
Psychological Bulletin (1990).
Thomas Gilovich, How We Know What Isn't So (Free Press, 1991). On the
asymmetric standards we apply to evidence: "Can I believe this?" for conclusions we
like, "Must I believe this?" for conclusions we do not.