Start Here: Data Basics

Question of 3In a table of orders, what does one row usually represent?

Each row is a single thing that happened, here one order. Columns describe it.

Question of 3A table has 500 rows and 12 columns. What does 12 tell you?

Columns are the details held about each row, so twelve describes each order in twelve ways.

Question of 3Your orders table has 1,000 rows. How many customers do you have?

One customer can place many orders, so rows and customers are rarely the same number.

Question of 3Which of these is most likely to be a column, not a row?

Order date describes a row. Orders and visits are things that happened, so they are rows.

Question of 3Where does a dashboard figure ultimately come from?

Every chart, figure and model sits on top of a table somewhere.

Question of 3In “orders by product category last quarter”, which is the measure?

Orders is the thing being counted. Category and quarter are how it is sliced.

Question of 3Which of these is a dimension?

Region splits the numbers. Revenue and average order value are measures.

Question of 3A colleague asks for “a report on customers”. What is missing?

Without a measure and dimensions, nobody knows what to count or how to break it down.

Question of 3“Average handling time by team, by week” contains how many dimensions?

Team and week are both dimensions. Average handling time is the measure.

Question of 3Which request is most useful to an analyst?

It names the measure, the dimensions and the period, so it can be built without guesswork.

Question of 3Nine customers spend £10 and one spends £500. Which figure best describes a typical customer?

The median ignores the extreme value, so it describes the typical customer better.

Question of 3Which situation makes an average most misleading?

Extreme values pull the average away from where most of the data sits.

Question of 3Two teams both average five days to resolve a ticket. What else do you need?

The same average can hide very different consistency, which is what customers feel.

Question of 3What does the median tell you?

The median is the middle value. The most common is the mode, and the last one is the mean.

Question of 3Your average order value jumps 40% in a month. What is the first thing to check?

A single unusually large order can move an average sharply, without anything else changing.

Question of 3Conversion goes from 10% to 12%. How should that be described?

The gap between two percentages is measured in percentage points. In relative terms it is a 20% increase.

Question of 3Which figure is missing from “conversion was 25% last week”?

Twenty-five per cent of four visits and of four thousand mean very different things.

Question of 3A campaign reports 100% growth in sign-ups. What should you check first?

Going from two to four is 100% growth, and usually not worth acting on.

Question of 3Which comparison removes seasonality best?

Year-on-year compares like with like, so Christmas is compared with Christmas.

Question of 3Market share rises from 4% to 6%. The correct phrasing is:

Two percentage points, which is also a 50% relative increase.

Question of 3Sales rose in the same week as a new advert. What is the strongest next step?

Other things change at the same time. A test tells you whether the advert caused it.

Question of 3Ice cream sales and sunburn both rise in July. What explains it?

A third factor driving both is the most common reason two things move together.

Question of 3Which gives the strongest evidence that a change caused an effect?

A test with a comparison group is the only one that isolates the change.

Question of 3Customers who use the mobile app spend more. What is the safest conclusion?

Self-selection is the usual explanation, and it is not the same as cause.

Question of 3What is the quickest sanity check on a causal claim?

Most false causal claims fall apart as soon as you list what else changed.

Question of 3You want to show how sales changed across twelve months. Which visual fits best?

A line chart shows change over time, and twelve points read at a glance.

Question of 3You are comparing revenue across eight regions. Which fits best?

Bars compare categories. A line would imply the regions follow an order.

Question of 3You want to see whether spend and satisfaction move together. Which fits best?

A scatter shows the relationship between two numeric measures.

Question of 3When is a pie chart a reasonable choice?

Pies work only for a few slices of a single whole, and rarely better than bars.

Question of 3A chart needs a paragraph of explanation. What does that usually mean?

If a chart needs explaining, the chart is doing the wrong job.

Question of 3A dashboard refreshes at 3am. You open it at 2pm. What does “today” include?

The figures are as fresh as the last refresh, so anything after 3am is not in there.

Question of 3Where is a sales figure first created?

The figure is born in the operational system, then travels.

Question of 3Your dashboard and the till system disagree by an hour’s trading. The likely reason is:

Timing differences explain most small gaps between a source system and a report.

Question of 3What is a data pipeline?

It is the plumbing that moves data from where it is created to where it is reported.

Question of 3Which question tells you most about a figure’s reliability?

Freshness and source are what decide whether you can act on it.

Question of 3Two teams report different customer numbers. What should you check first?

Different definitions explain most disagreements, so compare definitions before assuming an error.

Question of 3Marketing counts 1,240 leads, sales counts 480. The most likely reason is:

Both can be right. They are counting different things and calling them the same word.

Question of 3What settles this kind of disagreement permanently?

An agreed, recorded definition stops the argument recurring every month.

Question of 3A single source of truth means:

It is about one authoritative source per measure, not one system for everything.

Question of 3Which question opens the conversation best?

It assumes both are right about something, which is usually true and keeps people talking.

Question of 3What should you read first on a dashboard you have not seen before?

The furniture tells you what the numbers cover. Without it, a figure means nothing.

Question of 3Two colleagues quote different figures from the same dashboard. The most likely cause is:

Filters are personal to the session, so two people can read the same page and see different numbers.

Question of 3Where is the figure the designer thought mattered most, usually?

The eye lands top left, so that is where the headline figure belongs.

Question of 3A dashboard says 'last 30 days'. Someone asks about this month. What do you say?

Rolling 30 days and calendar month rarely match, especially early in a month.

Question of 3What does the refresh time tell you?

It tells you the point in time the figures describe.

Question of 3A bar chart's axis starts at 90 instead of 0. What is the effect?

Bar length is the message, so cutting the axis exaggerates every difference.

Question of 3Which chart type may legitimately start its axis away from zero?

Lines show movement rather than magnitude, so a tighter axis is often clearer.

Question of 3The last point on a monthly line drops sharply. What should you check first?

A part-month always looks like a crash. Cut it or mark it as partial.

Question of 3A chart shows this month only, with no comparison. What is missing?

A figure with nothing to compare it to cannot be judged as good or bad.

Question of 3Bars are sorted alphabetically. Why might that matter?

Sorting by value puts the answer at the top. Alphabetical hides it.

Question of 3Target is 100 and actual is 92. What is the variance?

Variance is actual minus target, so 92 minus 100 is minus 8.

Question of 3Costs come in 5% under budget. Is that good or bad?

Direction depends on the measure. Under budget on cost is usually good.

Question of 3Why can a percentage variance mislead?

Two out of four is a 50% miss, and may not matter at all.

Question of 3Which pair tells you most about performance?

An actual on its own cannot be judged. The comparison is the point.

Question of 3A report shows variance in both £ and %. Why show both?

A big percentage on a small base is a small problem, and the reverse is also true.

Question of 3Which request is easiest to answer?

It names the measure, the slices, the period and the comparison.

Question of 3What should a good question start from?

If no decision depends on it, the answer will not be used.

Question of 3You cannot say what you would do differently depending on the answer. That means:

It is the clearest sign that the question needs more thought first.

Question of 3Which is missing from 'revenue by region this month'?

Without last month, last year or a target, nobody can judge the number.

Question of 3Why does naming the period matter?

Period definitions are one of the most common causes of mismatched numbers.

Question of 3A satisfaction score comes from 12 responses. What should you ask?

Both the size and who is missing decide whether it means anything.

Question of 3Which is more reliable?

Larger samples move less from one period to the next.

Question of 3Only your happiest customers answer a survey. That is:

Who is missing shapes the answer as much as who replied.

Question of 3A weekly figure jumps around wildly. The most likely reason is:

Small numbers produce big percentage swings on their own.

Question of 3What is the simplest guard against over-reading a small sample?

Showing the count lets the reader judge the percentage.

Question of 3Your regional totals look wrong. What is the first thing to check?

Inconsistent naming silently splits one category into several.

Question of 3Totals in your extract do not match the source system. That means:

A mismatch is a question to answer, not a rounding difference to accept.

Question of 3Where should a data problem be fixed?

Fixing your copy leaves the same error for everyone else.

Question of 3Which of these is a timeliness problem?

Timeliness is about whether the data is fresh enough for the decision.

Question of 3You find duplicate rows. What is the risk?

Duplicates inflate counts and sums without any warning.

Question of 3Which of these is personal data?

An identifier that points to one person counts, even without a name.

Question of 3You remove names from a report about a three-person team. Is it anonymous?

Small groups identify people through context alone.

Question of 3What is the safest default when collecting personal data?

Less collected means less to protect and less to go wrong.

Question of 3Who should be able to see personal data?

Access follows purpose, not seniority or curiosity.

Question of 3Free-text comments in a survey are:

People name themselves, their team or their manager in comments all the time.