WEBVTT

00:00:07.808 --> 00:00:10.839
A toothpaste brand claims
their product will destroy more plaque

00:00:10.839 --> 00:00:12.910
than any product ever made.

00:00:12.910 --> 00:00:16.411
A politician tells you their plan
will create the most jobs.

00:00:16.411 --> 00:00:18.951
We're so used to hearing these
kinds of exaggerations

00:00:18.951 --> 00:00:20.850
in advertising and politics

00:00:20.850 --> 00:00:23.131
that we might not even bat an eye.

00:00:23.131 --> 00:00:26.111
But what about when the claim
is accompanied by a graph?

00:00:26.111 --> 00:00:28.471
Afterall, a graph isn't an opinion.

00:00:28.471 --> 00:00:32.611
It represents cold, hard numbers,
and who can argue with those?

00:00:32.611 --> 00:00:36.403
Yet, as it turns out, there are plenty
of ways graphs can mislead

00:00:36.403 --> 00:00:38.192
and outright manipulate.

00:00:38.192 --> 00:00:40.745
Here are some things to look out for.

00:00:40.745 --> 00:00:45.760
In this 1992 ad, Chevy claimed to make
the most reliable trucks in America

00:00:45.760 --> 00:00:47.510
using this graph.

00:00:47.510 --> 00:00:51.963
Not only does it show that 98% of all
Chevy trucks sold in the last ten years

00:00:51.963 --> 00:00:53.592
are still on the road,

00:00:53.592 --> 00:00:57.338
but it looks like they're twice
as dependable as Toyota trucks.

00:00:57.338 --> 00:01:00.634
That is, until you take a closer look
at the numbers on the left

00:01:00.634 --> 00:01:05.472
and see that the figure for Toyota
is about 96.5%.

00:01:05.472 --> 00:01:09.313
The scale only goes between 95 and 100%.

00:01:09.313 --> 00:01:12.963
If it went from 0 to 100,
it would look like this.

00:01:12.963 --> 00:01:16.243
This is one of the most common
ways graphs misrepresent data,

00:01:16.243 --> 00:01:18.333
by distorting the scale.

00:01:18.333 --> 00:01:20.804
Zooming in on a small portion
of the y-axis

00:01:20.804 --> 00:01:25.703
exaggerates a barely detectable difference
between the things being compared.

00:01:25.703 --> 00:01:27.974
And it's especially misleading 
with bar graphs

00:01:27.974 --> 00:01:31.023
since we assume the difference
in the size of the bars

00:01:31.023 --> 00:01:33.233
is proportional to the values.

00:01:33.233 --> 00:01:36.125
But the scale can also be distorted
along the x-axis,

00:01:36.125 --> 00:01:40.414
usually in line graphs
showing something changing over time.

00:01:40.414 --> 00:01:44.747
This chart showing the rise
in American unemployment from 2008 to 2010

00:01:44.747 --> 00:01:47.996
manipulates the x-axis in two ways.

00:01:47.996 --> 00:01:50.395
First of all, the scale is inconsistent,

00:01:50.395 --> 00:01:53.416
compressing the 15-month span
after March 2009

00:01:53.416 --> 00:01:56.755
to look shorter than 
the preceding six months.

00:01:56.755 --> 00:02:00.106
Using more consistent data points
gives a different picture

00:02:00.106 --> 00:02:03.705
with job losses tapering off
by the end of 2009.

00:02:03.705 --> 00:02:06.675
And if you wonder why
they were increasing in the first place,

00:02:06.675 --> 00:02:10.615
the timeline starts immediately after
the U.S.'s biggest financial collapse

00:02:10.615 --> 00:02:12.626
since the Great Depression.

00:02:12.626 --> 00:02:15.219
These techniques are known as
cherry picking.

00:02:15.219 --> 00:02:18.869
A time range can be carefully chosen
to exclude the impact of a major event

00:02:18.869 --> 00:02:20.648
right outside it.

00:02:20.648 --> 00:02:24.762
And picking specific data points
can hide important changes in between.

00:02:24.762 --> 00:02:27.356
Even when there's nothing wrong
with the graph itself,

00:02:27.356 --> 00:02:30.937
leaving out relevant data can give
a misleading impression.

00:02:30.937 --> 00:02:33.997
This chart of how many people watch
the Super Bowl each year

00:02:33.997 --> 00:02:37.626
makes it look like the event's
popularity is exploding.

00:02:37.626 --> 00:02:40.198
But it's not accounting 
for population growth.

00:02:40.198 --> 00:02:41.967
The ratings have actually held steady

00:02:41.967 --> 00:02:45.109
because while the number 
of football fans has increased,

00:02:45.109 --> 00:02:47.959
their share of overall viewership has not.

00:02:47.959 --> 00:02:49.888
Finally, a graph can't tell you much

00:02:49.888 --> 00:02:53.318
if you don't know the full significance 
of what's being presented.

00:02:53.318 --> 00:02:56.457
Both of the following graphs
use the same ocean temperature data

00:02:56.457 --> 00:02:59.719
from the National Centers 
for Environmental Information.

00:02:59.719 --> 00:03:02.490
So why do they seem to give
opposite impressions?

00:03:02.490 --> 00:03:05.279
The first graph plots the average
annual ocean temperature

00:03:05.279 --> 00:03:07.987
from 1880 to 2016,

00:03:07.987 --> 00:03:10.149
making the change look insignificant.

00:03:10.149 --> 00:03:12.878
But in fact, a rise of even
half a degree Celsius

00:03:12.878 --> 00:03:15.799
can cause massive ecological disruption.

00:03:15.799 --> 00:03:17.219
This is why the second graph,

00:03:17.219 --> 00:03:19.858
which show the average temperature
variation each year,

00:03:19.858 --> 00:03:22.390
is far more significant.

00:03:22.390 --> 00:03:27.379
When they're used well, graphs can
help us intuitively grasp complex data.

00:03:27.379 --> 00:03:31.180
But as visual software has enabled
more usage of graphs throughout all media,

00:03:31.180 --> 00:03:35.900
it's also made them easier to use
in a careless or dishonest way.

00:03:35.900 --> 00:03:39.560
So the next time you see a graph,
don't be swayed by the lines and curves.

00:03:39.560 --> 00:03:40.882
Look at the labels,

00:03:40.882 --> 00:03:42.130
the numbers,

00:03:42.130 --> 00:03:43.048
the scale,

00:03:43.048 --> 00:03:44.360
and the context,

00:03:44.360 --> 00:03:46.780
and ask what story the picture
is trying to tell.