Scatterplots and regression

A scatterplot, also called a scattergraph or scatter diagram, is a plot of

the data points in a set. It plots data that takes two variables into account
at the same time. Here are some examples of scatterplots:

20 20

15 15

10 10

5 5

0 0
5 10 15 20 5 10 15 20

20 20

15 15

10 10

5 5

0 0
5 10 15 20 5 10 15 20

Even though scatterplots can look like a mess, sometimes we’re able to
see trends in the data. For example, the two graphs on the left definitely
seem to be roughly following a line: the one on top looks like it follows a
line with a positive slope; the bottom one looks like it follows a line with a
negative slope.

353
The graph in the upper right looks like it might be following a positively-
sloped line, but if it is, the trend is not as clear as either of the graphs on
the left.

And the graph in the lower right doesn’t look like it’s following any trend at
all.

When we say that the data in a scatterplot appears to follow a trend, what
we’re really saying is that it appears to follow some line, or maybe some
other kind of curve, like for example an exponential curve or sinusoidal
curve. No matter the shape of the curve that the data follows, we call it the
approximating curve, and the process of finding the equation of the
approximating curve is called curve fitting.

The regression line is one of the most important approximating curves

we’ll talk about, so let’s take a look at that now.

Regression line
It was intuitive for us to start looking for trends in the scatterplots as soon
as we saw the plotted points. And, in fact, spotting trends is probably
what we spend most of our time doing when we work with scatterplots.
The plot alone isn’t super helpful, but if we can use the plot to observe
some kind of a trend in the data, then we might be able to use that trend
to draw conclusions or make predictions about the data.

The most common way that we’ll do this is with a regression line. It’s the
line that best shows the trend in the data given in a scatterplot. A

354
regression line is also called the best-fit line, line of best fit, or least-
squares line.

The regression line is a trend line we use to model a linear trend that we
see in a scatterplot, but realize that some data will show a relationship that
isn’t necessarily linear. For example, the relationship might follow the
curve of a parabola, in which case the regression curve would be parabolic
in nature. For the rest of this lesson we’ll focus mostly on linear regression.

Equation of the regression line

There are a few ways to calculate the equation of the regression line. The
equation for a regression line is most often given in slope-intercept form,
ŷ = a + bx, where x is the independent, or explanatory variable, b is the
slope, and a is the intercept when x = 0, or the y-intercept. Sometimes we’ll
see the regression equation written as ŷ = β0 + β1x, but this is exactly the
same thing since β0 + β1 is the same as a + b. The regression line formula
then calculates the slope b and the y-intercept a using

n ∑ xy − ∑ x ∑ y
b=
n ∑ x 2 − ( ∑ x)2

∑ y − b∑ x
a=
n

In the formulas for the slope b and y-intercept a,

n is the number of data points in the set,

∑
xy is the sum of all the products of the x and y,

355
 ∑
x is the sum of all the x-values,

∑
y is the sum of all the y-values,

x 2 is the sum of all the squared x-values, and

∑

x)2 is the square of the sum of all the x-values.

∑
(

Once we find the equation of the regression line, we denote it with y,̂
(pronounced “y-hat”), to indicate that it’s a regression line, and remind us
that it’s an approximation for the data set. So the equation of the
regression line is

ŷ = a + bx

Let’s work through an example of how to find the equation of the

regression line.

Example

Find the least-squares line for the data set.

356
 x y

0 0.8

2 1.0

4 0.2

6 0.2

8 2.0

10 0.8

12 0.6

We’ll start by calculating the slope, m. There are 7 data points in this set,
so n = 7. It can be helpful to calculate xy and x 2 for each data point, plus
find the sum of the x-values and the sum of the y-values, and add all of
these into the data table, since we’ll be using them in our calculations. Our
new table that includes this extra information will be

x y xy x2

0 0.8 0 0

2 1.0 2 4

4 0.2 0.8 16

6 0.2 1.2 36

8 2.0 16 64

10 0.8 8 100

12 0.6 7.2 144

Sum: 42 5.6 35.2 364

357
Let’s plug what we’ve found into the formula for slope.

n ∑ xy − ∑ x ∑ y
b=
n ∑ x 2 − ( ∑ x)2

7(35.2) − (42)(5.6)
b=
7(364) − (42)2

246.4 − 235.2
b=
2,548 − 1,764

11.2
b=
784

b ≈ 0.0143

Now let’s plug what we’ve found into the formula for the y-intercept.

∑ y − b∑ x
a=
n
11.2
5.6 − 784
(42)
a=
7

5.6 − 0.6
a=
7

5
a=
7

a ≈ 0.7143

In statistics, we usually write the slope and y-intercept to the ten-

thousandths place (four decimal places) to prevent severe rounding errors

358
from occurring in the estimated values, and giving us an inaccurate
regression line. Therefore, we can say that the regression line is given
approximately by

ŷ = a + bx

ŷ = 0.0143x + 0.7143

Let’s plot the data points on a scatterplot and then add in the regression
line we found to double-check ourselves.

2.0

1.6

1.2

0.8

0.4

0
2 4 6 8 10 12

The regression line looks like it runs roughly through the data, indicating
the trend.

With this last example, we might notice that the data actually wasn’t super
linear. If we look at the scatterplot we made, we might even say it has

359
more of a sinusoidal shape, and we can see that the point around x = 8
looks like an outlier.

So the next question we need to answer starts to become obvious, and

that is “Is the regression line even a good estimate of the data?” Luckily,
there are ways to measure how good of a fit this line is to the data points,
and we’ll look at those techniques a little bit later on.

Describing the trend

Whenever we describe a relationship in the data, we should describe

• the form (linear, parabolic, sinusoidal, etc.),

• the direction (positive, negative),

• the strength (strong, weak), and

• the outliers (outliers, no outliers).

Form

If the data roughly follows a linear trend line, we can say the relationship is
linear. If the data more closely follows a parabolic curve, we would say the
relationship is parabolic. If the scatterplot just looks like one big blob, and
we can’t really see any relationship in the data, then we would say there’s
no relationship or correlation at all.

Linear correlation:

360
 20

15

10

-5
5 10 15 20

Parabolic correlation:

30

22.5

15

7.5

0
3.75 7.5 11.25 15

No correlation:

20

15

10

0
5 10 15 20

Direction

361
If the regression line has a positive slope, the data has a positive linear
relationship; if the regression line of the data has a negative slope, the
data has a negative linear relationship.

Positive linear relationship:

20

15

10

-5
5 10 15 20

Negative linear relationship:

20

15

10

-5
5 10 15 20

Strength

If the data is clustered tightly around its regression line, we might say it
shows a strong linear relationship. If the data is loosely clustered, we
might say it shows a moderate linear relationship. A weak linear
relationship would be data that is spread out but still noticeably in the
form of a trend line or curve.

362
 Strong linear relationship:

20

15

10

-5
5 10 15 20

Moderate linear relationship:

20

15

10

0
5 10 15 20

Outliers

Whether the data has a strong or weak relationship of any kind can also
be affected by the existence of outliers, or lack thereof. Remember that an
outlier is a data point that lies far away from the trend line.

363
 Outlier

20

15

10

0
5 10 15 20

If all of the data points are very tightly clustered, then there are no
outliers, which means the data shows a strong relationship. But if there are
some or many outliers away from the majority, then the data shows a
moderate relationship.

The more outliers there are, and the further away they are, the weaker the
relationship. The fewer outliers there are, and the more tightly clustered
the data, the stronger the relationship.

Example

Describe any trend in the data, in terms of form, direction, strength, and
outliers.

364
 20

15

10

0
5 10 15 20

Let’s look at each part one at a time: form, direction, strength, and
outliers.

Form: The scatterplot appears to have a roughly linear relationship,

as opposed to a parabolic, or other relationship.

Direction: It certainly has a positive relationship, because the data

moves up and to the right.

Strength and outliers: The strength of the relationship is moderate,

which is due to how spread out the data points are, and the
existence of outliers in the data set, like (6,18).

If we plot the trend line, we get

365
 20

15

10

0
5 10 15 20

If we take out a few points that are further from the regression line, like
(6,18), (11,4), (14,8), and (17,20),

20

15

10

0
5 10 15 20

we can see that the new, adjusted regression line fits the remaining data a
little bit better:

366
 20

15

10

0
5 10 15 20

If we plot both lines on the original scatterplot, we get

20

15

10

0
5 10 15 20

and we can see the effect that some of these outliers have on the
regression line.

367
The purpose of regression
So what’s the purpose of curve fitting in general, or finding the equation of
the regression line specifically? Well, the main purpose for finding the
approximating curve, whether it’s a regression line or a regression curve
with some other shape, is to come up with an equation that we can use to
make predictions.

In the first example from this section, we were given a data table:

x y

0 0.8

2 1.0

4 0.2

6 0.2

8 2.0

10 0.8

12 0.6

If we were asked to give an approximate value of y for x = 9, or for x = 100,

based on this data set, it’d be awfully hard to do using just the data points
in the table. After all, the table doesn’t give a value for x = 9, and it
certainly doesn’t give a value for x = 100.

But if we calculate the equation of the regression line that approximates

the data, then we can plug x = 9, x = 100, or any other value into the
equation, and we’ll get back an estimated value of y.

368
And that’s the purpose of regression. Technically, regression is just the
process of estimating the value of the dependent variable from a given
value of the independent variable.

369