Confounding
variables are a type of variable

related to a model's independent
and dependent variables.

A variable must meet two conditions
to be a confounder.

One, it must be correlated
with the independent variable.

And two, it must be causally related
to the dependent variable.

A pretty classic example

tasks you with collecting data on sunburns
and ice cream consumptions.

If we find that ice cream consumption

goes up at the same time
as instances of sunburn,

are we meant to conclude
that ice cream causes sunburn?

One missing variable here is temperature
as a proxy for the amount of sun

as a confounding variable.

The hotter
it is, the more likely people are to eat

ice cream, as well as get a sunburn
as they spend more time outdoors.

One of the challenges of modeling is
that it is necessary to do an exhaustive

search for possible confounding factors,
as their absence could lead to

certain algorithms to detect relationships
that don't actually exist.

By the same token,
if the confounding factor is identified,

but not properly handled
by the algorithm in question,

they could end up
exaggerating relationships that do exist.