Missing Is Not Zero

The most consequential cleaning decision, and the one most often made by accident.

Replacing missing with 0 changes every average, sum and rate you compute afterwards. Dropping rows with missing values silently changes who is in your analysis, which is worse, because the number still looks fine.

Decide per column, and write it down. For each column with gaps: what does missing mean here, what will you do about it, and how many rows are affected?

Then report the count. 'Excluded 214 rows with no recorded year, 4.8% of the total' is a sentence a reader can evaluate. Silence is not.

✦ Sentinel values are real

0, -1, 999 and 9999 are all used as 'no data' markers in real datasets, and none of them announce themselves. Look at the distribution of a column before trusting it: a suspicious spike at exactly 999 is not a coincidence, and averaging it in produces a number that is confidently wrong.