One panel per region, and one region recorded no rows this month - what should the panel set show, and what is the danger?
answer
- empty panel against no panel at all
- declared set of groups, or inferred
- a complete-looking page asks nothing
- zeros are a measurement, emptiness is not
- enumerate the groups you expect
basics
~20 sTwo outcomes are possible: a panel with axes and no marks, or no panel at all. The second is dangerous: the page looks complete, and nothing distinguishes a region with no rows from a region that does not exist.
solid answer
~50 sWhich outcome you get depends on where the chart learned its list of groups. If the grouping column declares its full set of possible values, a value nobody observed still produces a panel - an empty one, with its label and axes and no marks. If the chart infers the groups from the values actually present, the region is never drawn and the page looks complete. The missing panel is the failure that matters: absence of evidence has been rendered as absence of the group, and nothing invites the reader to ask. The fix is to build the panel set from the groups you expect rather than from the rows that arrived, and to let the empty panel say which kind of empty it is. Do not fill it with zeros to even up the grid - an empty panel says nothing was recorded, while zeros claim a measurement that came out zero.
go deeper
Recall that a group with no rows may appear as an empty panel or may not appear at all, and that a page which simply omits it looks complete when it is not.
Explain what decides it: whether the set of groups is declared up front or worked out from the values present in the rows that arrived.
Show that you build the panel set from the groups you expect and annotate the empty ones, so a broken feed shows up as a labelled hole rather than as a tidy grid.
The rule worth setting is that any recurring report enumerates its groups from a reference list, because the reports that fail quietly are the ones whose scope is defined by their own input.
## Two different pages for the same data A panel per region is drawn, and one region contributed no rows at all this month. There are exactly two things the page can look like: - **An empty panel.** The region has a panel, with its label and its axes, and no marks inside it. A reader sees the region, sees it is empty, and can ask why. - **No panel.** The grid has one fewer panel. Nothing on the page refers to that region in any way. A reader sees a full, tidy grid and asks nothing. These pages carry different claims about the world, and the difference is not a matter of styling. The second page is a stronger and less supportable claim than the first, made silently. ## Where the panel set comes from Which page you get is decided by something upstream of the chart: whether the set of groups is **declared** or **inferred**. - Where the grouping column carries a declared set of possible values, a value that no row happens to take is still part of the set. It occupies a slot, and it produces a panel - usually an empty one. - Where the chart works out the groups from the values it can see in the rows, a value that never appears has no existence as far as the chart is concerned. It cannot be drawn, because nothing told the chart it was possible. That is a genuine difference in mechanism between designs, and it is why no default can be quoted here. The consequence for you is the same either way: **if the panel set is derived from the rows that arrived, the chart can only ever be as complete as the data was.** ## Why the missing panel is the dangerous case | what the page shows | what the reader concludes | what it may actually mean | |---|---|---| | a panel with no marks | this group exists and recorded nothing | no rows collected, or rows whose values were all absent | | a panel full of zeros | this group was measured and the answer was zero | a genuine zero, or an invented value filling a gap | | no panel at all | this group is not part of the picture | the group had no rows, or the group was never in scope | The first row is a question the reader can ask. The third row is a question the reader never knows to ask, and that is the whole of the difference. The failures that matter in reporting are rarely wrong numbers on the page - they are true numbers surrounded by an absence nobody noticed. A region that stopped reporting because a feed broke looks, on the inferred page, exactly like a region that was never on the report. ## What an empty panel should say An empty panel is better than a missing one, but an empty panel that says only *nothing here* is still weak. Distinguish, in the panel or in a note next to it: - **no rows were received** for this group in this period - a collection fact; - **rows were received and the values were all absent** - a data fact about the column; - **the measured value was zero** - which is not an empty panel at all, and must be drawn as a zero rather than as emptiness. Filling an empty panel with zeros to even up the grid collapses the first two cases into the third and invents a measurement. It is the most common well-intentioned version of this mistake, and it is worse than the missing panel, because it replaces an absence with a confident false value. ## Making the page robust 1. **Derive the panel set from the groups you expect**, not from the groups that turned up. The expected list may come from a reference list of regions, from the declared values of the grouping column, or from the previous period's set - but it comes from somewhere other than this month's rows. 2. **Draw the panel even when it is empty**, so the absence occupies space and has a label. Space on the page is what makes a reader ask. 3. **Annotate the empty panel with the reason**, at whatever level of detail you have. Even *no data received* is far more than an empty frame conveys on its own. 4. **Check what the empty panel does to the limits.** Under one shared range an empty panel still draws its axes, so the hole is visible and framed; where each panel is fitted to its own rows, a panel with no rows may have no limits to fit and can render as a blank rectangle or not render at all. ## What the interviewer is testing Whether you treat the panel set as a **choice** rather than as a property of the data. A candidate who says the chart shows what is in the data has described the failure without noticing it. The chart shows what somebody decided it should enumerate, and deciding that from this month's rows is how an outage becomes an empty space nobody sees.
- The tool draws only the groups it saw in the rows. How do you make the absence visible?Supply the group list yourself instead of letting the rows define it, and draw a panel for every expected group so the empty ones take up space and carry a label. Then annotate each empty panel with what is known - no rows received, or values all absent. The point is that the hole occupies a position on the page.
- An empty panel and a panel of zeros sit side by side. What does each claim?The empty panel claims nothing was recorded for that group. The panel of zeros claims a measurement was taken and its answer was zero. Drawing zeros in place of an empty panel therefore invents data, and it is the more damaging of the two errors because it looks like a finding rather than a gap.
- Does the choice of shared or per-panel limits change how an empty panel behaves?Yes. With one shared range the empty panel still draws the same axes as every other, so the gap is framed and obvious. Where each panel is fitted to its own rows there may be no rows to fit limits to, and the panel can come out as a blank rectangle or fail to be laid out at all, which makes the absence harder to see rather than easier.
saying these in an interview costs you the question
- Assumes a group with no rows will always appear as an empty panel.
- Reads a group's absence from the page as the group having had no activity.
- Fills the empty panel with zeros so that the grid looks even.
- Thinks the panel set is a property of the data rather than a choice.
- Treats a missing panel as cosmetic rather than as a silent claim.