Regression, Predicting Numbers
A property site quotes the house on Maple Street at $412,000, and no agent typed that figure. A line drawn through thousands of past sales took the home's size and read the price straight off it.
- ▸The best line touches almost none of the houses; it just sits closest to all of them at once.
- ▸A straight line that runs low in the middle, high at the ends is the wrong shape, not bad luck.
- ▸Push a prediction past the last size you measured and it stops being a reading, it turns into a guess.
A property site has two questions about the same house on Maple Street. First: what will it sell for? The honest answer is a number on a sliding scale, maybe $380,000, maybe $412,000, and being a few thousand off still counts as close. Second: will it sell this month, yes or no? That answer is one of a small fixed set of buckets, with no halfway.
That first kind of question, the one whose answer is a quantity on a scale, is regression. Predicting a price, a temperature, a number of clicks, all regression: the output can land anywhere in a range, and how wrong a guess is depends on how far off it lands, not on right versus wrong.
This whole topic lives inside that first question. Everything ahead, drawing the line that turns a house's size into a predicted price and reading a number straight off it, sits on top of one job: answer with a value on a scale, not a bucket.