0:00:06.000,0:00:08.760
Confounding[br]variables are a type of variable

0:00:09.000,0:00:12.880
related to a model's independent[br]and dependent variables.

0:00:13.360,0:00:16.800
A variable must meet two conditions[br]to be a confounder.

0:00:17.320,0:00:21.280
One, it must be correlated[br]with the independent variable.

0:00:21.640,0:00:26.560
And two, it must be causally related[br]to the dependent variable.

0:00:27.280,0:00:29.040
A pretty classic example

0:00:29.040,0:00:34.520
tasks you with collecting data on sunburns[br]and ice cream consumptions.

0:00:34.520,0:00:37.320
If we find that ice cream consumption

0:00:37.320,0:00:40.920
goes up at the same time[br]as instances of sunburn,

0:00:41.240,0:00:44.860
are we meant to conclude[br]that ice cream causes sunburn?

0:00:45.520,0:00:50.680
One missing variable here is temperature[br]as a proxy for the amount of sun

0:00:50.680,0:00:52.760
as a confounding variable.

0:00:52.760,0:00:56.200
The hotter[br]it is, the more likely people are to eat

0:00:56.200,0:01:01.070
ice cream, as well as get a sunburn[br]as they spend more time outdoors.

0:01:01.280,0:01:06.160
One of the challenges of modeling is[br]that it is necessary to do an exhaustive

0:01:06.160,0:01:10.960
search for possible confounding factors,[br]as their absence could lead to

0:01:10.960,0:01:15.730
certain algorithms to detect relationships[br]that don't actually exist.

0:01:15.920,0:01:19.920
By the same token,[br]if the confounding factor is identified,

0:01:20.160,0:01:23.490
but not properly handled[br]by the algorithm in question,

0:01:23.600,0:01:27.960
they could end up[br]exaggerating relationships that do exist.