Missing Data vs Zero: An Empty Cell Against a Real Nothing

A blank says nobody wrote anything down. A zero says something happened and it came to nothing. In the Neelbagh stall record, invented, NB-05 Bansi Flour's month 2 cell is blank and NB-04 Peetal Utensils' month 2 cell is a true zero. Reading either as the other moves the headline by Rs 1,733.33/- one way and Rs 1,935.59/- the other.

Two things are already in hand before any of this, and neither is rebuilt below. The arithmetic averageThe figures being counted are totalled, and that total is split between however many were counted. Nothing else enters into it, and the operation was set out in full earlier. is already established here, and so is the convention that governs how this record applies it. So are the reasons a cell can turn up empty, and which divisor belongs to which question, each covered separately. One pair of cells, a blank and a measured zero, is worked to the bottom below.

One convention decides every figure printed below, so it goes down before any of them. The average takingsWhat a stall rang up over a month, measured at the point of sale. Nothing has been taken off yet for the pitch, for help behind the table or for buying the goods, so it is turnover and never earnings. per stall month is the sum of the takings cells that carry a usable number, divided by how many carry one. The sum of the usable cells is the numeratorThe figure sitting above the line in a division, the one being shared out. In an average it is the total of everything that was added up. and the count of them is the denominatorThe figure sitting below the line in a division, the one the total is shared between. In an average it is simply how many things were divided by.. Three consequences follow, and the third carries the whole argument below. An empty cell feeds neither half. The code the office writes when no return reaches it feeds neither half either. A measured zero feeds a row into the count while feeding nothing into the total. Any figure below that divides by something else names its divisor in the same breath.

Cleaned of the faults that belong to other treatments, the Neelbagh stall record leaves twenty nine usable takings figures adding to Rs 15,71,700/-. Under the convention those twenty nine average Rs 54,196.55/- per stall month. Every figure below is arithmetic on those twenty nine numbers, and the entire argument is about one cell that is inside them and one cell that is not.

What does a blank cell say, and what does a zero say?

A blank and a zero look almost the same on a printed sheet. One column, two rows, one cell with nothing in it and one cell with a small round mark in it. A tired eye slides over both. And they are not near neighbours, not shades of the same idea, and not two ways of writing the same fact.

A blank is a statement about the record. The blank says nobody wrote anything in the cell. The blank says nothing beyond that, and what it does say is a fact about paperwork: about a clerk, a form, a keyboard, a moment when somebody was interrupted. A blank carries no information whatsoever about what happened out in the world.

A zero is a statement about the world. The zero says the takings were looked at, and the answer was nothing. Somebody counted and reached none. A zero is a measurement, and a measurement of nothing is still a measurement.

Take the everyday version. Picturing it costs nothing. A bank passbook, opened at a week with no entries printed on it. Does that week show that nothing was spent? It does not. The empty week might mean nothing was spent, and it might mean the branch had not printed up to date, and it might mean the payments were made in cash. An empty line in a passbook is a fact about the passbook. Set that beside a shopkeeper who says plainly that he opened on Tuesday, stood there all day and sold nothing. The shopkeeper's sentence is a fact about Tuesday. The passbook line and the shopkeeper's sentence look equally like nothing, and only one of them is about the world.

Which two cells in the Neelbagh stall record are which?

Month 2 of the Neelbagh stall record, invented, is where both cells sit. The market office keeps one row for each stall in each month. NB-03 Harit Greens filed twice, so month 2 runs to nine rows for eight stalls. Rows that repeat are covered separately and are left alone here.

Read down the takings column and stop at the fifth and sixth rows. NB-04 Peetal Utensils reads Rs 0/-. NB-05 Bansi Flour reads nothing at all. The two cells sit one directly under the other, in the same column, in the same month, and everything that follows rests on their being nothing like the same thing.

How is NB-04 Peetal Utensils' zero known to be genuine?

Because somebody went and found out, and the finding is written down where anybody can go and read it. NB-04 Peetal Utensils sells steel plates, tumblers and cooking pots. In month 2 its pitchA numbered square of ground the market lets out by the month. Nothing is built on it, so whoever takes the square arrives with their own table, awning and goods. was being re roofed, so the stall did not open on a single day of the month. The stall kept its licence. NB-04 Peetal Utensils filed the return the market office asks for, and on that return the takings line reads nothing.

So there are three separate facts here, and it is worth separating them. The stall existed. Its trading was observed. The observation was that it traded nothing. NB-04 Peetal Utensils' zero is not a gap in the record, it is a finding, and a finding of nothing belongs in every count the record supports.

The whole of the second fault, further down, comes from people not believing that a finding of nothing belongs in a count. In an average of what a stall in this market takes in a month, a stall that took nothing is one of the months. Left out, the figure is no longer an average of what a stall takes. The figure becomes an average of what a stall takes in the months it takes something, and nobody asked that question.

How is NB-05 Bansi Flour's blank known not to be a zero?

Because the figure exists. NB-05 Bansi Flour sells atta, rice and dal, and it was open every day of month 2. NB-05 Bansi Flour filed a paper returnThe printed form a trader fills in by hand at the end of a month and hands to the office, kept afterwards in a drawer. A document, not a row in a file. like every other stall, and that paper return reads Rs 52,000/-. Somebody in the market office then typed the month into the file and did not type that one cell.

Compare the two documents side by side and the difference is not subtle. NB-04 Peetal Utensils filed a return slipThe short one line form the office accepts from a stall that did not trade at all in a month, in place of the full return. It states that fact and nothing else. stating that it traded on no day. NB-05 Bansi Flour filed a full return carrying a figure. One document says nothing happened. The other document says something happened and here is how much, and the record simply failed to copy it.

So NB-05 Bansi Flour's empty cell says exactly nothing about the market and something quite specific about the market office. The true figure for that cell is not lost, not unknowable and not a matter of judgement. The figure is sitting in a drawer eight steps from the desk where the file was typed.

What does writing a zero into the blank do to the figure?

Now cost it. Somebody has to put something in NB-05 Bansi Flour's empty cell before the month can be totalled, and they have two candidates: a zero, or the Rs 52,000/- on the paper return. Fill it either way and the record holds thirty usable figures rather than twenty nine, so the count is thirty in both cases.

A zero adds nothing to a sum, so writing a zero leaves the numerator at Rs 15,71,700/-. The average takings per stall month over those thirty figures then comes to Rs 52,390/-. Write the Rs 52,000/- the paper return actually shows and the numerator becomes Rs 16,23,700/-, so the same average over the same thirty figures comes to Rs 54,123.33/-. The distance between the two readings is Rs 1,733.33/-.

Notice what moved and what did not: the count is thirty in both readings, so writing the zero is purely a fault in the numerator, and it is short by exactly the Rs 52,000/- that nobody typed. The fault is worth naming plainly. A zero written into an empty cell is not a cautious choice or a neutral placeholder. A written zero is an assertion, and the assertion is that a stall which opened every day of the month took nothing across the counter. Nobody observed that. Nobody wrote it down. The written zero arrives in the record in exactly the same typeface as every measured figure beside it, and from that moment on nothing distinguishes it from a fact.

What does dropping a genuine zero do to the figure?

Now the fault running the other way. The same confusion starts it, and it arrives somewhere else entirely. Somebody totalling the Neelbagh stall record sees NB-04 Peetal Utensils' Rs 0/-, decides a zero is not really a figure, and leaves the row out of the count.

Under the convention as it stands, twenty nine usable figures add to Rs 15,71,700/- and average Rs 54,196.55/- per stall month. Drop NB-04 Peetal Utensils' zero and the sum is still Rs 15,71,700/-. A zero was never adding anything to it. The count, though, falls to twenty eight. The same money over twenty eight figures averages Rs 56,132.14/-, a reading Rs 1,935.59/- higher than it should be.

So this one is the mirror image: the numerator is held at Rs 15,71,700/- in both readings and only the denominator moves, by exactly one row. Of the two faults it is the quieter, and the quieter one earns the more attention. A program that reads every empty cell as a zero is a bug, and a bug leaves fingerprints: somebody eventually spots a stall reporting nothing, asks why, and finds the setting. A person quietly skipping zeros while adding up by hand leaves nothing at all. No setting, no log, no odd looking row. The figure simply comes out higher and no one can say why.

Which part of the average does each fault attack?

Set the two costings next to each other and a clean split shows up, and it is the most portable thing here. Both faults come from the same confusion, and they damage opposite halves of the same division.

Reading NB-05 Bansi Flour's blank as a zero attacks the numerator. Thirty figures are counted either way; the sum is short by the Rs 52,000/- that was never copied out of the drawer. Reading NB-04 Peetal Utensils' zero as though the cell were empty attacks the denominator. A zero contributes nothing to a sum whether it is kept or not, so the sum is Rs 15,71,700/- either way. The count is short by one row.

An empty cell mishandled costs money out of the total, and a measured nothing mishandled costs a row out of the count, and once a cell can be named as one or the other, the half of the arithmetic at risk is known. The split is why the distinction is worth this much attention. The distinction is not a matter of tidiness. Naming a cell decides where the damage lands.

What do the four readings look like side by side?

The whole argument as a table. Every row divides some sum by some count, every row names its own divisor in the row, and the four are built from the same invented Neelbagh stall record with one cell handled four different ways.

Two comparisons live in that table and each isolates one fault. The first row against the second holds the count at thirty and moves Rs 52,000/- of numerator, a gap of Rs 1,733.33/-. The third row against the fourth holds the sum at Rs 15,71,700/- and moves one row of denominator, a gap of Rs 1,935.59/-. Every one of the four readings is arithmetically correct, and three of them answer a question nobody asked. Rs 54,123.33/- is the only row where both cells say what their documents say, and so the only row that describes what the eight stalls of the Neelbagh market actually took.

If one fault pulls down and the other pulls up, do they cancel?

Everybody asks that at this point, and the instinct behind the question is reasonable. One fault costs Rs 1,733.33/-, the other costs Rs 1,935.59/-, they push in opposite directions, and a record carrying both ought to come out roughly right. So work it rather than guess it.

Take the Neelbagh stall record with both faults in place. NB-05 Bansi Flour's blank has been filled with a zero, adding a row worth nothing to the count. NB-04 Peetal Utensils' measured zero has been dropped, removing a row worth nothing from the count. The sum never budges from Rs 15,71,700/-, one row went out and one came in, the count settles at twenty nine, and the record reports Rs 54,196.55/-. Rs 54,196.55/- is the same figure the convention gives, and the two are equal for a plain arithmetic reason and not for an interesting one: identical sum, identical count.

Now compare it with the truth. Handled properly, this record holds thirty figures adding to Rs 16,23,700/-, and thirty into that sum is Rs 54,123.33/-. So the record carrying both faults reports Rs 73.22/- above the truth. Two faults worth over Rs 1,700/- each have left a visible trace of under Rs 100/-. The pair is far worse than either fault on its own because the only signal there was is now gone.

So do they cancel? No, and here is the proof rather than the assurance. Work out where that Rs 73.22/- comes from and it turns out to be one thirtieth of the distance between the record's own reading of Rs 54,196.55/- and what NB-05 Bansi Flour really took, Rs 52,000/-. The distance is Rs 2,196.55/-, and Rs 2,196.55/- shared over thirty rows is Rs 73.22/-. The near miss is therefore a property of one figure in this record. NB-05 Bansi Flour happened to have a month close to the record's own average, so the residue came out small. Nothing about two opposite faults produced it, and had that stall taken a very different figure the same two faults would have left the record much further adrift.

How is it decided, standing in front of an empty cell?

Everything above collapses into one question, and it is a sentence rather than a rule. Can the reason this cell is empty be named?

If the answer is that the stall traded and took nothing, a zero goes in. Writing that zero is recording a measurement. If the answer is that nobody typed it in, the cell stays empty until somebody fetches the figure. The figure exists and somebody has it. And if the answer is that nobody knows, then the cell stays empty and the not knowing gets written down beside it. Writing a zero because a zero is easier to compute with is the single commonest way a made up number enters a record, and it enters wearing the same clothes as every measured figure around it.

Empty cells are not a market office problem. A lender pulling a small trader's monthly turnover sees blanks in the months the trader was hospitalised and blanks in the months the accountant was slow, and the two look identical on the statement. An analyst averaging a set of monthly figures finds a run of empty cells and has to decide, cell by cell, whether they mean nothing sold or nothing recorded. A household adding up a year of spending finds three months with no entries and has to remember whether that was a quiet quarter or a lost notebook. In every one of those, the question is the same and the answer comes from outside the file: from a day bookThe office's running log, kept in ordinary prose rather than in rows. Somebody adds a line whenever something worth remembering happens, and getting a figure back out of prose has its own treatment., a drawer, a phone call, a memory. The record itself cannot say.