The Chart That Chose Its Own Scale
A chart that starts its axis conveniently, aggregates away the one bad week, or smooths a line that should be flat and noisy. The honest limits of visual presentation, and what a number needs before it is worth drawing.
A chart is a set of choices made before a single value is drawn. Where the axis starts. What is grouped with what. Whether a line is smoothed. Every one of those choices is legitimate in isolation, and every one of them can make a bad week disappear while leaving the numbers untouched.
The trouble is that the choices are invisible by the time the reader looks. A reader sees a shape, and the shape already contains a decision about which days to group and where to put the floor. So the chart is often not a picture of the data. It is a picture of one reading of the data, presented with the authority of a picture.
Three ways a chart decides for you
What a chart can do that the underlying numbers will not
It can choose the floor
A line chart can start its axis at any point. Starting above the data makes small movements look like a collapse. Starting at zero makes the same movements look like nothing. Both are technically correct renderings of the same values, and the reader cannot tell from the picture which one they are looking at unless the axis is labelled.
It can aggregate away the bad week
Grouping days into weeks is a normal thing to do and a reliable way to hide one bad day. A week containing four ordinary days and one collapse looks like a slightly weaker week. The collapse is not in the picture at all, and the total has not changed, so nothing in the underlying records says the week went wrong.
It can smooth a line that should be noisy
A moving average or a trend line replaces every point with an average of its neighbours. This makes real variation look like gentle drift, and it makes a genuine break look like a curve. The smoothed value is not wrong, and it is also not what happened on any particular day.
None of these is a lie. That is what makes them worth understanding. A person who fudges a number is doing something you can, in principle, catch. A person who picks a reasonable aggregation and a reasonable axis has produced a chart that is true about the group and silent about the part, and there is nothing to catch.
The practical consequence is that a chart should never be the place where a conclusion is first reached. It is a good place to end up after the numbers are already on the page, because then the reader knows what to look at and the chart helps rather than leads.
The week that looked fine
A worked example, constructed for this article
Every figure below is invented to show the shape of the problem. None of it is a real business and none of it is a benchmark. The arithmetic is shown so you can check it by hand.
Collections for ten recorded days:
| Day | Collected |
|---|---|
| 1 | 41,200 |
| 2 | 38,700 |
| 3 | 44,900 |
| 4 | 39,400 |
| 5 | 7,600 |
| 6 | 42,100 |
| 7 | 40,300 |
| 8 | 43,500 |
| 9 | 39,800 |
| 10 | 41,700 |
The ten-day total is 3,79,200. Check it by adding in pairs:
- 41,200 + 38,700 = 79,900
- 44,900 + 39,400 = 84,300
- 7,600 + 42,100 = 49,700
- 40,300 + 43,500 = 83,800
- 39,800 + 41,700 = 81,500
- 79,900 + 84,300 = 1,64,200
- 1,64,200 + 49,700 = 2,13,900
- 2,13,900 + 83,800 = 2,97,700
- 2,97,700 + 81,500 = 3,79,200
Now the weekly chart. Group the same ten days into two groups of five and the bad day vanishes:
- Week 1, days 1 to 5: 1,71,800
- Week 2, days 6 to 10: 2,07,400
- 2,07,400 - 1,71,800 = 35,600
The chart says week 2 was 35,600 better than week 1. What actually happened is that day 5 collected 7,600, and an ordinary day in this series is close to 40,000. The shortfall on that one day is:
- 40,000 - 7,600 = 32,400
So the picture of a normal fortnight improving by 35,600 is, underneath, one day on which something went wrong and a grouping decision that made sure nobody could see it. The total did not change. What changed is that the reader was shown a number too coarse to contain the problem.
The axis, on the same data. Take four of the ordinary days: 38,700, 39,400, 43,500 and 41,700. The spread across them is:
- 43,500 - 38,700 = 4,800
Draw those four on an axis starting at 35,000 and they look like a mountain range. Draw the identical four values on an axis starting at zero and they look like a flat line. Same four numbers. The reader's conclusion about whether collections are volatile or steady is being produced by the floor of the axis, not by the business.
The smoothing, on the same data. A three-day average across days 4, 5 and 6 is:
- 39,400 + 7,600 + 42,100 = 89,100
- 89,100 / 3 = 29,700
For comparison, days 3 and 4 together average:
- 44,900 + 39,400 = 84,300; 84,300 / 2 = 42,150
So the smoothed line does dip, by 42,150 - 29,700 = 12,450. But it dips to a number that never existed, on a day that was not like that, and it dips gently enough to read as noise. The single day that deserves an explanation is now a shallow bend. That is the specific thing smoothing does: it converts an event into a trend.
Three conclusions, and none of them requires the chart to be badly made. The grouping hid one bad day. The axis floor manufactured a trend out of a range of 4,800. The smoothing turned a day that lost 32,400 against an ordinary day into a curve that bends. In each case a reasonable person, given reasonable data, arrived at a wrong impression, and no amount of care about the source records would have prevented it.
The other thing the example shows is that nothing here required the underlying records to be bad. Every one of the ten values is a real transaction total from the constructed series. The distortion happened entirely in the presentation layer, which is exactly why it is so hard to catch: the export is correct, the report is correct, and only the picture is misleading.
What a number needs before it is worth drawing
Checklist: is this number ready to be a picture?
- The period is stated as dates, and the day count is known. Without a day count, a gap and a quiet week look identical.
- The basis is named: invoiced, collected, outstanding or quoted. A line that mixes bases is not a series.
- The gaps are understood. Any day with no recorded transaction is a question to answer before it is drawn, because a gap silently distorts every average computed across it.
- The grouping is chosen deliberately and stated. Weekly, monthly, or daily, and the reason, written where the reader can see it.
- The axis floor is stated, and if the axis is truncated the caption says so.
- No smoothing, no trend line, and no interpolation across a gap, unless the method is named and the raw points are also shown.
- The raw values are available alongside the picture. A chart a reader cannot check against the numbers is an assertion with decoration.
That last item is the one that would have prevented every failure in the example above. If the table of ten daily values sits on the same page as the picture, a reader who suspects the aggregation can check it in thirty seconds, and the picture stops being authoritative and starts being helpful.
The awkward truth is that most of the checklist has to be satisfied before a chart is drawn, not after. Choosing a grouping and a floor is part of deciding what the number means, which is the same work that has to be done for a table anyway. Charts are not a shortcut around that work. They are a presentation of its result, and they inherit whatever it got wrong.
The honest limits of visual presentation
Charts are genuinely good at some things and it is worth saying what, because the argument here is not against them. A shape that took an hour to establish in a table becomes obvious in a second: a level shift, a step change, a run of similar values, a divergence between two series. The comparison of ten daily figures to two weekly aggregates is exactly the kind of thing a picture does better than a column of numbers.
What a chart cannot do is show you the thing you were not looking for. The bad day in the example is invisible precisely because it is unusual, and unusual is the category worth surfacing. A chart compresses toward the typical and spends its resolution on the range. That is a good trade for spotting a trend and a bad one for spotting an exception, and exceptions are usually why anybody opens a report.
The honest way round it is to use the picture for the trend and the table for the exceptions, and to be explicit about which is which. A chart labelled with its period, its grouping, its floor and its basis, sitting next to the values, is a legitimate presentation. A chart with a headline that states a conclusion, drawn on a truncated axis, showing a smoothed line, at a weekly granularity, in a deck that will be read on a phone, is a device for controlling what a reader concludes, and it is usually used to support a conclusion that was chosen first.
It is worth being clear about what that means for the pages that do describe views. The day-end view is built from the transactions of the day: shifts, payment mix, expected cash, counted cash, variance and exceptions. The owner-level view is built from what was quoted, what was billed, what was collected and what is outstanding. Those are views over records rather than assembled figures, and a view over records can be walked back to the records that produced it, which is the property that makes a number defensible in a meeting.
In practice, on Monday morning
The practical version of this article is a short routine. Print the daily values, not only the weekly or monthly aggregate. Look for the day that is not like the others and treat it as a question rather than as noise. When a comparison is shown to somebody else, say the period, the grouping and the floor out loud, because those three sentences are the difference between a picture and an argument.
And when a chart is put in front of you, by us or by anyone else, three questions are enough. What period, at what granularity, starting from where? If the answers come back clean, the picture is worth something. If they come back vague, the picture is not telling you about the business, it is telling you about the presentation, and the numbers underneath are the only thing worth reading.
A related failure sits close to this one and is worth naming. A figure that returns a different value on the second export of the same afternoon is not a presentation problem at all, and no amount of careful charting will fix it. That is covered in the article on a report that changes when you look twice. The failure here is different in an important way: the numbers are stable and correct, and only the picture of them has been arranged.
Frequently asked questions
Is starting a chart axis above zero dishonest?
No, and it is a normal choice. It is misleading only when the reader cannot see it. A truncated axis makes small movements look dramatic, and the identical values on a zero-based axis look flat. The honest move is to label the floor and say that the axis is truncated, so the reader can judge the movement rather than the picture.
How much detail can a chart hide through aggregation?
A great deal, and the worked example in this article shows it. Grouping ten daily values into two groups of five hid a day that collected 7,600 against an ordinary day near 40,000, a shortfall of 32,400, behind a weekly picture that said the second week was 35,600 better. The totals never changed; only the resolution of the picture did.
Does smoothing a line make it dishonest?
Not by itself, but it replaces every point with an average of its neighbours, which converts an event into a trend. A day that lost 32,400 becomes a gentle bend through a value that never occurred on any day. If smoothing is used, the method should be named and the raw points shown alongside it.
Do you have a BI tool or a custom report builder?
No. There is no business intelligence tool, no dashboard builder, no scheduled report delivery, no data-warehouse sync and no configurable custom report builder. Reporting is a set of standard views over the records, and a bespoke report or a chart of your own data is something we would scope rather than something you configure.
Does this article make any claim about tax or compliance?
No. The 18% GST rate appears in the other articles in this series only to keep a constructed invoice total checkable by hand. Treatment of any real invoice belongs with your own chartered accountant, and NoxOrigin does not file GST returns or any other statutory return.